Embed Size (px)
Citation preview
DOI 101167tvst452
Article
Restricted Summed-Area Projection for GeographicAtrophy Visualization in SD-OCT Images
Qiang Chen1 Sijie Niu1 Honglie Shen1 Theodore Leng3 Luis de Sisternes2 andDaniel L Rubin2
1 School of Computer Science and Engineering Nanjing University of Science and Technology Nanjing China2 Department of Radiology and Medicine (Biomedical Informatics Research) Stanford University School of Medicine Stanford CA USA3 Byers Eye Institute at Stanford Stanford University School of Medicine Palo Alto CA USA
Correspondence Qiang ChenSchool of Computer Science andEngineering Nanjing University ofScience and Technology 200 XiaoLing Wei Nanjing 210094 Chinachen2qiangnjusteducn
Received 14 January 2015Accepted 21 June 2015Published September 2015
Keywords geographic atrophy vi-sualization spectral domain opti-cal coherence tomographyrestricted summed-area projectionchoroidal vasculature geographicseparability
Citation Chen Q Niu S Shen H LengT de Sisternes L and Rubin DLRestricted summed-area projectionfor geographic atrophy visualizationin SD-OCT images 20154(5)2doi101167tvst452
Purpose To enhance the rapid assessment of geographic atrophy (GA) across themacula in a single projection image generated from three-dimensional (3D) spectral-domain optical coherence tomography (SD-OCT) scans by introducing a novelrestricted summed-area projection (RSAP) technique
Methods We describe a novel en face GA visualization technique the RSAP byrestricting the axial projection of SD-OCT images to the regions beneath the Bruchrsquosmembrane (BM) boundary and also considering the choroidal vasculaturersquos influenceon GA visualization The technique analyzes the intensity distribution beneath theretinal pigment epithelium (RPE) layer to fit a cross-sectional surface in the sub-RPEregion The area is taken as the primary GA projection A median filter is then adoptedto smooth the generated GA projection image The RSAP technique was evaluated in99 3D SD-OCT data sets from 27 eyes of 21 patients presenting with advancednonexudative age-related macular degeneration and GA We used the meandifference between GA and background regions and GA separability metric tomeasure GA contrast and distinction in the generated images respectively Wecompared our results with two existing GA projection techniques the summed-voxelprojection (SVP) and Sub-RPE Slab techniques
Results Comparative results demonstrate that the RSAP technique is more effectivein displaying GA than the SVP and Sub-RPE Slab The average of the mean differencebetween GA and background regions and the GA separability based on SVP Sub-RPESlab and RSAP were 01290880 02380919 and 02760938 respectively
Conclusions The RSAP technique was more effective for GA visualization than theconventional SVP and Sub-RPE Slab techniques Our technique decreases choroidalvasculature influence on GA projection images by analyzing the intensity distributioncharacteristics in sub-RPE regions The generated GA projection image with the RSAPtechnique has improved contrast and distinction
Translational Relevance Our method for automated generation of GA projectionimages from SD-OCT images may improve the visualization of the macularabnormalities and the management of GA
Introduction
Age-related macular degeneration (AMD) is the
leading cause of blindness among the elderly Ad-
vanced AMD where vision loss is mostly noticed is
most common in a nonexudative form characterized by
the presence of geographic atrophy (GA)1ndash3 Here
atrophy refers to the degeneration of the deepest cells
of the retina To date the primary retinal layer affectedby the evolution of GA remains unclear GArepresents the loss of either photoreceptors (PR)retinal pigment epithelium (RPE) or choriocapillarislayers within the macula4ndash6 However most histopath-ological studies suggest that the initial event in GAoccurrence is RPE cell loss followed with ensuing PRcell death and choriocapillaris atrophy7ndash9
In the recent years spectral-domain optical coher-
1 TVST j 2015 j Vol 4 j No 5 j Article 2
1
ence tomography (SD-OCT) has become the prefer-able imaging method to acquire high-speed high-resolution high-density three-dimensional (3D) im-ages covering the central macula10 and has provensuccessful in identifying the GA11 Bearelly et al6
studied the PR-RPE interface in GA using SD-OCTto test in vivo to determine if SD-OCT provides anadequate resolution for reproducible measurement ofthe PR layer at the margins of GA Fleckenstein etal1213 used SD-OCT to describe a wide spectrum ofmorphologic alterations that appear within theatrophic area as well as within the surroundingretinal tissue Traditionally progression and enlarge-ment of GA has been assessed on fundus autofluo-rescence images (FAF)14 FAF has remained the goldstandard for direct GA visualization as it demarcatesthe lesion borders more precisely than clinicalobservation on fundus photography taking advantageof the optical properties of lipofuscin and photore-ceptor degeneration products15 Schmitz-Valckenberget al16 compared the FAF and OCT appearance ofeyes with GA and showed that the mean length of anatrophic lesion measured on the FAF image had theclosest agreement with the appearance of choroidalhyperreflectivity on the OCT B-scan and the reduc-tion of the FAF signal seen from GA was spatiallycorrelated with the abrupt transition on the SD-OCTB-scan from a hyporeflective choroid to a hyper-reflective choroid Sayegh et al17 concluded that SD-OCT seems to be an appropriate imaging modality forevaluating the extent of GA lesions and the area ofchoroidal signal enhancement on the SD-OCT B-scans correlated well with the hypofluorescence areameasured on FAF Color fundus photographs (CFPs)can also be used for GA visualization and assessmenthowever it is difficult to make reproducible quanti-tative measurements of GA in CFPs18
SD-OCT imaging can produce a fundus image(OFI) for visualizing GA by generating an en facesummed-voxel projection (SVP) of all the B-scans619ndash21 However the SVP fundus images may not ideal forGA assessment as other retinal pathologies with highreflectivity such as drusen may interfere with GAvisualization Stopa et al22 overcame this problem byidentifying pathologic retinal features with colormarkings in each B-scan image before the imagevolume was collapsed along the depth axis to producethe SVP While effective this technique is timeconsuming because of the human interaction requiredA newer technique introduced into OCT imagingdevices is the Sub-RPE Slab which creates an en faceimage only from the light reflected from beneath the
RPE The Sub-RPE Slab has higher contrast at theborders of GA than the conventional SVP techniqueby excluding the highly reflective retinal layers abovethe RPE23 To increase the contrast further the Sub-RPE Slab technique was improved by combiningchoroidal brightening and RPE thinning24 HoweverRPE thinning is not always seen in all areas of GA soit can only be used as an auxiliary characteristic forGA visualization Stetson et al25 recently presented anOCT minimum intensity (MI) projection to predictlocations of growth at the margin of GA and thegrowth rate outside the margin The MI image issuitable for observing the margin for GA growth butis limited by low contrast between GA regions andbackground regions
Although the Sub-RPE Slab technique can gener-ate an en face GA projection image with bettercontrast than the Stetson technique and the SVPfundus image it does not consider choroidal vascu-lature which has low reflectivity and results indecreased contrast within GA regions In this paperwe present a new fundus projection method calledrestricted summed-area projection (RSAP) whichimproves GA visualization by considering the contri-bution of the choroidal vasculature to a GAprojection image
Materials and Methods
Anatomically the choroid can be divided into twoprincipal components the choriocapillaris a lobularvascular plexus comprised of large fenestrated capil-laries adjacent to Bruchrsquos membrane (BM) and thechoroidal stroma26 The main principle of the existingfundus projection GA visualization techniques gener-ated from SD-OCT images lies in the identification ofthe typical choroidal brightening that appears in theregions affected by GA However the many bloodvessels in the choroid which normally manifest in lowreflection values decrease the contrast and distinctionof macular regions affected by GA As an exampleFigure 1 shows the choroidal vasculature influence onGA visualization where Figures 1A and 1B are theSVP and Sub-RPE Slab projection images generatedfrom one 3D SD-OCT scan acquired with a CirrusOCT (Carl Zeiss Meditec Inc Dublin CA) systemrespectively The RPE boundaries were delineated byhand Figure 1C displays the B-scan corresponding tothe yellow dashed line in Figure 1A and severalstructures that can be observed in this image aremanually labeled and shown in Figure 1D We canobserve that the contrast (difference between bright
2 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Figure 1 Choroidal vasculature influence on GA visualization (A) SVP projection image (B) Sub-RPE Slab projection image (C) ExampleB-scan (D) Example B-scan with structure labels (E) Sub-CSI Slab projection image
3 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
regions where GA is present and darker backgroundregions) of the Sub-RPE Slab projection image isbetter than that of the SVP projection image becausethe Sub-RPE Slab projection excludes the highlyreflective layers above RPE such as the retinal nervefiber layer (RNFL) and ellipsoid zone (EZ) whichmay influence visualization as they are added onto theprojection image Due to the presence of thechoroidal vasculature within the region affected byGA we can observe dark regions within the GAextent in both of the projection images as markedwith the orange circle in Figure 1A
In order to overcome the influence of the choroidalvasculature on GA visualization one could followtwo different strategies (1) excluding the choroidalvasculature with a new Sub- Choroidal-Sclera Inter-face (CSI) Slab technique similar to that done withthe Sub-RPE Slab technique but restricting theprojection region beneath the CSI boundary or (2)detecting and filling the dark regions where choroidalvessels are present with high intensity values With thefirst strategy the choroidal vasculature would becompletely excluded because no vessels appearbeneath the CSI boundary but the main character-istic of GA in SD-OCT imaging choroidal brighten-ing would also be omitted As less light penetratesthrough the choroid the brightening differencebetween the regions inside and outside GA wouldalso be less dramatic reducing contrast An exampleillustrating this effect is shown in Figure 1E where theimage produced by the Sub-CSI Slab projectiontechnique shows less contrast for GA visualizationthan the earlier proposed techniques (Figs 1A 1B)The CSI boundaries for Sub-CSI Slab projection weredelineated manually The Sub-RPE (Fig 1B) andSub-CSI (Fig 1E) projections are to generate the enface image by collapsing the pixels beneath the RPEand CSI respectively It does not mean pixel intensityis only plotted along the identified boundaries (RPEor CSI) The pixels in the narrow band beneath theRPE or CSI are collapsed to generate the projectionimages In this work we focus on the second proposedstrategy forming an image based in the projection ofintensity values restricted to the choroid region whilealso detecting and filling with high intensity values thedarker regions within the extent of GA that arecaused by vessel presence We call this technique therestricted summed-area projection (RSAP)
GA Visualization Based on RSAP
The main strategy of the RSAP technique lies inusing the intensity distribution beneath the RPE layer
as observed in SD-OCT images to fill the low intensityregions produced by the presence of choroidal vesselsVessel presence is identified by analyzing intensityprofiles An example of intensity distribution beneaththe RPE in a typical SD-OCT scan namely the regionbetween the two dashed green curves in Figure 1Cwhere GA is present is displayed in Figure 2 We canobserve that the intensity distribution decreasestoward the x direction (depth) and the rate of thisdecrease in the GA region is typically slower than thatin the normal region Figures 3A and 3B show twointensity profiles (marked with the red curve) of thesub-RPE region (Fig 2A) in the GA and the normalregions respectively The large concave regionpointed out in Figure 3A corresponds to the choroidalvessel marked with the orange oval in Figure 1D Thisvessel would have a great influence on GA visualiza-tion that is solely based on intensity summation suchas with the SVP and Sub-RPE Slab methods It can beseen from Figure 3 that for the same x position in thechoroidal region (axial location) the intensity valuesin the GA region (Fig 3A) are larger than those in thenormal region (Fig 3B) except in particular axiallocations (x axis) where choroidal vessel regions arelocated (marked with the orange oval circle in Fig 1Dand the dashed black circle in Fig 3A) In otherwords regions where GA is present will have higherintensity profiles than those where GA is not presentwhile vessel location can be identified with thepresence of local minimums in such profiles Thisprinciple constitutes the basic idea of the proposedRSAP technique
The flowchart of our technique is shown in Figure4 which comprises the following operations
(1) Segmentation of BM boundary the BM bound-aries were segmented with the automatic 3Dgraph search method27
(2) Flattening of sub-RPE region an image iscomposed by taking the recorded intensity valuesbeneath the segmented BM boundary up to amaximum depth where GA can be detectedconstituting a flattened sub-RPE image Thismaximum depth is set up as an independentparameter in this work (depth of sub-RPE asexplained in the later parameter evaluation) Theflattened sub-RPE region is shown in Figure 2A
(3) Finding local maximum intensity points For eachcolumn in the flattened sub-RPE region (A-scanlocation) the points with local maximum inten-sity value (namely the intensity value of the pointis larger than those of its two connected points)
4 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
are found as marked with the blue circle inFigure 3
(4) Locating maximum intensity points at higherdepths The maximum intensity points whosevalue follows a constantly decreasing functionwith depth (x axis) are selected as marked withthe magenta stars in Figure 3 The purpose of thisstep is to ensure a constantly descendent intensityprofile beneath RPE
(5) Calculating the area below the surface construct-ed with the maximum intensity points at higherdepths We interpolate the intensity profile in the
axial locations between the selected maximumintensity points at higher depths using linearinterpolation (magenta lines in Fig 3) andcalculate the area of the polygon formed by thisinterpolation and a baseline of zero intensity (areabelow the magenta lines marked in Fig 3)
(6) Taking the calculated area above as the primaryGA projection value at each projection locationand
(7) Using a median filter to smooth the generated GAprojection image to alleviate the noise influenceand make the final GA projection image smoother
Figure 2 Intensity distribution analysis (A) Flattened sub-RPE region (B) Intensity surface of (A)
5 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
a simple median filter with a 3 3 3 neighborhood
was used
Figure 5A shows the GA projection image with the
proposed RSAP technique and Figure 5B shows a
detail comparison of three GA projection techniques
in regions of interest corresponding to the red dashedrectangles in Figure 6A Compared with the SVP andSub-RPE Slab projection images (Figs 1A 1B) theRSAP projection image displays a higher contrast andalso overcomes the influence of the choroidal vascu-lature on GA visualization as shown in Figure 5B
Figure 3 Intensity profiles of columns 100 (A) and 370 (B) marked with the dashed blue and yellow lines in Figure 2A respectively
Red curve intensity values of two columns Blue circle local maximum intensity point Magenta star maximum intensity point at higher
depths (in the x direction) (A) Intensity distribution in GA region (B) Intensity distribution in normal region
Figure 4 Flowchart of the proposed technique
6 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Evaluation Study
To evaluate the RSAP technique for GA visualiza-tion 99 3D SD-OCTmacular images from 27 eyes of 21patients presenting with advanced nonexudative AMDwere collected and analyzed The data was acquiredlongitudinally with several observations from the sameeye as a GA was followed up in regular clinicalpractice All eyes were previously diagnosed with GA
and collected consecutively from clinical practice The
study protocol was approved by an institutional review
board and the HIPAA compliant research adhered to
the Declaration of Helsinki and all federal and state
laws Each 3D OCT data set was acquired over a 63 6
mm area (corresponding to 512 3 128 pixels) with a
1024-pixel (2 mm) axial resolution on a commercial SD-
OCT device (Cirrus OCT Carl Zeiss Meditec Inc)
Figure 5 GA projection image with RSAP (A) RSAP projection image (B) Results by the three discussed techniques in regions of interest
as indicated in (A)
Figure 6 Depth of sub-RPE region for GA separability for one SD-OCT scan Four GA projection images corresponding to four red points
are inset
7 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Both qualitative analysis and quantitative evalua-tions were performed on the data For the quantita-tive evaluation we computed two metrics to measurethe GA contrast and distinction
Mean difference (MD) between GA and back-ground regions defined by
MD frac14lGA lbg
lGA thorn lbg
eth1THORN
where lGA and lbg represent the average intensity valuesin GA and background regions respectively The GAand background regions are outlined manually in GAprojection imagesAn examplemanual annotationof theGAboundaries is shown in the top-left image in Figure 6markedwithagreenoutlineWeused theMDtomeasureGA contrast against background regions Higher MDvalues corresponded to higher contrast levels
GA separability of the GA projection image wasdefined as
SGA frac14 1max254
ifrac141
XK
jfrac141
jBiethjTHORN GethjTHORNj
Keth2THORN
where Bi indicates the binarized result with a globalthreshold of value i with the GA projection imagenormalized to the interval [0 255] and G indicates aGA segmentation result outlined manually K denotesthe number of the pixel in the GA projection image InBi andG the GA and background regions are binarizedto 1 and 0 respectively The reasoning behind Equation2 is to calculate the maximum overlap ratio between allof the binarization results generated with the global
threshold and a lsquolsquogold standardrsquorsquo outlined in the SVPprojection images by an expert ophthalmic grader Therange of separability values is from 0 to 1 The best GAseparability (ie SGA frac14 1) represents the GA regionsthat can be completely separated from the backgroundby a global threshold namely the binarization resultsgenerated with the global threshold and the lsquolsquogoldstandardrsquorsquo are the same This metric was used tomeasure the GA distinction Since one purpose ofgenerating GA projection images is to reliably deter-mine GA extent28 this separability measurementreflects the feasibility in the generated images
Results
The proposed RSAP technique was tested andcompared with the SVP and Sub-RPE Slab tech-niques qualitatively and quantitatively The influenceof the parameter controlling the maximum considereddepth for the sub-RPE region mentioned earlier wasfirst evaluated and later fixed for all 99 test images
Parameter Evaluation
Figure 6 shows the GA separability for differentdepths of sub-RPE regions in one SD-OCT imagewhere the depth was varied from 100 to 300 pixels(approximately 02~06 mm in the collected images)with an interval of 10 pixels Figure 6 demonstrates thatthe GA separability increases when the depth increasesfrom 100 to 240 pixels and then remains stable withfurther depth increases Figure 7 shows the relationship
Figure 7 Mean (redline) and standard deviation (pale pink shading) of the GA separability across the 99 SD-OCT images in this study
considering different values of depth of sub-RPE region
8 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
ence tomography (SD-OCT) has become the prefer-able imaging method to acquire high-speed high-resolution high-density three-dimensional (3D) im-ages covering the central macula10 and has provensuccessful in identifying the GA11 Bearelly et al6
studied the PR-RPE interface in GA using SD-OCTto test in vivo to determine if SD-OCT provides anadequate resolution for reproducible measurement ofthe PR layer at the margins of GA Fleckenstein etal1213 used SD-OCT to describe a wide spectrum ofmorphologic alterations that appear within theatrophic area as well as within the surroundingretinal tissue Traditionally progression and enlarge-ment of GA has been assessed on fundus autofluo-rescence images (FAF)14 FAF has remained the goldstandard for direct GA visualization as it demarcatesthe lesion borders more precisely than clinicalobservation on fundus photography taking advantageof the optical properties of lipofuscin and photore-ceptor degeneration products15 Schmitz-Valckenberget al16 compared the FAF and OCT appearance ofeyes with GA and showed that the mean length of anatrophic lesion measured on the FAF image had theclosest agreement with the appearance of choroidalhyperreflectivity on the OCT B-scan and the reduc-tion of the FAF signal seen from GA was spatiallycorrelated with the abrupt transition on the SD-OCTB-scan from a hyporeflective choroid to a hyper-reflective choroid Sayegh et al17 concluded that SD-OCT seems to be an appropriate imaging modality forevaluating the extent of GA lesions and the area ofchoroidal signal enhancement on the SD-OCT B-scans correlated well with the hypofluorescence areameasured on FAF Color fundus photographs (CFPs)can also be used for GA visualization and assessmenthowever it is difficult to make reproducible quanti-tative measurements of GA in CFPs18
SD-OCT imaging can produce a fundus image(OFI) for visualizing GA by generating an en facesummed-voxel projection (SVP) of all the B-scans619ndash21 However the SVP fundus images may not ideal forGA assessment as other retinal pathologies with highreflectivity such as drusen may interfere with GAvisualization Stopa et al22 overcame this problem byidentifying pathologic retinal features with colormarkings in each B-scan image before the imagevolume was collapsed along the depth axis to producethe SVP While effective this technique is timeconsuming because of the human interaction requiredA newer technique introduced into OCT imagingdevices is the Sub-RPE Slab which creates an en faceimage only from the light reflected from beneath the
RPE The Sub-RPE Slab has higher contrast at theborders of GA than the conventional SVP techniqueby excluding the highly reflective retinal layers abovethe RPE23 To increase the contrast further the Sub-RPE Slab technique was improved by combiningchoroidal brightening and RPE thinning24 HoweverRPE thinning is not always seen in all areas of GA soit can only be used as an auxiliary characteristic forGA visualization Stetson et al25 recently presented anOCT minimum intensity (MI) projection to predictlocations of growth at the margin of GA and thegrowth rate outside the margin The MI image issuitable for observing the margin for GA growth butis limited by low contrast between GA regions andbackground regions
Although the Sub-RPE Slab technique can gener-ate an en face GA projection image with bettercontrast than the Stetson technique and the SVPfundus image it does not consider choroidal vascu-lature which has low reflectivity and results indecreased contrast within GA regions In this paperwe present a new fundus projection method calledrestricted summed-area projection (RSAP) whichimproves GA visualization by considering the contri-bution of the choroidal vasculature to a GAprojection image
Materials and Methods
Anatomically the choroid can be divided into twoprincipal components the choriocapillaris a lobularvascular plexus comprised of large fenestrated capil-laries adjacent to Bruchrsquos membrane (BM) and thechoroidal stroma26 The main principle of the existingfundus projection GA visualization techniques gener-ated from SD-OCT images lies in the identification ofthe typical choroidal brightening that appears in theregions affected by GA However the many bloodvessels in the choroid which normally manifest in lowreflection values decrease the contrast and distinctionof macular regions affected by GA As an exampleFigure 1 shows the choroidal vasculature influence onGA visualization where Figures 1A and 1B are theSVP and Sub-RPE Slab projection images generatedfrom one 3D SD-OCT scan acquired with a CirrusOCT (Carl Zeiss Meditec Inc Dublin CA) systemrespectively The RPE boundaries were delineated byhand Figure 1C displays the B-scan corresponding tothe yellow dashed line in Figure 1A and severalstructures that can be observed in this image aremanually labeled and shown in Figure 1D We canobserve that the contrast (difference between bright
2 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Figure 1 Choroidal vasculature influence on GA visualization (A) SVP projection image (B) Sub-RPE Slab projection image (C) ExampleB-scan (D) Example B-scan with structure labels (E) Sub-CSI Slab projection image
3 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
regions where GA is present and darker backgroundregions) of the Sub-RPE Slab projection image isbetter than that of the SVP projection image becausethe Sub-RPE Slab projection excludes the highlyreflective layers above RPE such as the retinal nervefiber layer (RNFL) and ellipsoid zone (EZ) whichmay influence visualization as they are added onto theprojection image Due to the presence of thechoroidal vasculature within the region affected byGA we can observe dark regions within the GAextent in both of the projection images as markedwith the orange circle in Figure 1A
In order to overcome the influence of the choroidalvasculature on GA visualization one could followtwo different strategies (1) excluding the choroidalvasculature with a new Sub- Choroidal-Sclera Inter-face (CSI) Slab technique similar to that done withthe Sub-RPE Slab technique but restricting theprojection region beneath the CSI boundary or (2)detecting and filling the dark regions where choroidalvessels are present with high intensity values With thefirst strategy the choroidal vasculature would becompletely excluded because no vessels appearbeneath the CSI boundary but the main character-istic of GA in SD-OCT imaging choroidal brighten-ing would also be omitted As less light penetratesthrough the choroid the brightening differencebetween the regions inside and outside GA wouldalso be less dramatic reducing contrast An exampleillustrating this effect is shown in Figure 1E where theimage produced by the Sub-CSI Slab projectiontechnique shows less contrast for GA visualizationthan the earlier proposed techniques (Figs 1A 1B)The CSI boundaries for Sub-CSI Slab projection weredelineated manually The Sub-RPE (Fig 1B) andSub-CSI (Fig 1E) projections are to generate the enface image by collapsing the pixels beneath the RPEand CSI respectively It does not mean pixel intensityis only plotted along the identified boundaries (RPEor CSI) The pixels in the narrow band beneath theRPE or CSI are collapsed to generate the projectionimages In this work we focus on the second proposedstrategy forming an image based in the projection ofintensity values restricted to the choroid region whilealso detecting and filling with high intensity values thedarker regions within the extent of GA that arecaused by vessel presence We call this technique therestricted summed-area projection (RSAP)
GA Visualization Based on RSAP
The main strategy of the RSAP technique lies inusing the intensity distribution beneath the RPE layer
as observed in SD-OCT images to fill the low intensityregions produced by the presence of choroidal vesselsVessel presence is identified by analyzing intensityprofiles An example of intensity distribution beneaththe RPE in a typical SD-OCT scan namely the regionbetween the two dashed green curves in Figure 1Cwhere GA is present is displayed in Figure 2 We canobserve that the intensity distribution decreasestoward the x direction (depth) and the rate of thisdecrease in the GA region is typically slower than thatin the normal region Figures 3A and 3B show twointensity profiles (marked with the red curve) of thesub-RPE region (Fig 2A) in the GA and the normalregions respectively The large concave regionpointed out in Figure 3A corresponds to the choroidalvessel marked with the orange oval in Figure 1D Thisvessel would have a great influence on GA visualiza-tion that is solely based on intensity summation suchas with the SVP and Sub-RPE Slab methods It can beseen from Figure 3 that for the same x position in thechoroidal region (axial location) the intensity valuesin the GA region (Fig 3A) are larger than those in thenormal region (Fig 3B) except in particular axiallocations (x axis) where choroidal vessel regions arelocated (marked with the orange oval circle in Fig 1Dand the dashed black circle in Fig 3A) In otherwords regions where GA is present will have higherintensity profiles than those where GA is not presentwhile vessel location can be identified with thepresence of local minimums in such profiles Thisprinciple constitutes the basic idea of the proposedRSAP technique
The flowchart of our technique is shown in Figure4 which comprises the following operations
(1) Segmentation of BM boundary the BM bound-aries were segmented with the automatic 3Dgraph search method27
(2) Flattening of sub-RPE region an image iscomposed by taking the recorded intensity valuesbeneath the segmented BM boundary up to amaximum depth where GA can be detectedconstituting a flattened sub-RPE image Thismaximum depth is set up as an independentparameter in this work (depth of sub-RPE asexplained in the later parameter evaluation) Theflattened sub-RPE region is shown in Figure 2A
(3) Finding local maximum intensity points For eachcolumn in the flattened sub-RPE region (A-scanlocation) the points with local maximum inten-sity value (namely the intensity value of the pointis larger than those of its two connected points)
4 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
are found as marked with the blue circle inFigure 3
(4) Locating maximum intensity points at higherdepths The maximum intensity points whosevalue follows a constantly decreasing functionwith depth (x axis) are selected as marked withthe magenta stars in Figure 3 The purpose of thisstep is to ensure a constantly descendent intensityprofile beneath RPE
(5) Calculating the area below the surface construct-ed with the maximum intensity points at higherdepths We interpolate the intensity profile in the
axial locations between the selected maximumintensity points at higher depths using linearinterpolation (magenta lines in Fig 3) andcalculate the area of the polygon formed by thisinterpolation and a baseline of zero intensity (areabelow the magenta lines marked in Fig 3)
(6) Taking the calculated area above as the primaryGA projection value at each projection locationand
(7) Using a median filter to smooth the generated GAprojection image to alleviate the noise influenceand make the final GA projection image smoother
Figure 2 Intensity distribution analysis (A) Flattened sub-RPE region (B) Intensity surface of (A)
5 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
a simple median filter with a 3 3 3 neighborhood
was used
Figure 5A shows the GA projection image with the
proposed RSAP technique and Figure 5B shows a
detail comparison of three GA projection techniques
in regions of interest corresponding to the red dashedrectangles in Figure 6A Compared with the SVP andSub-RPE Slab projection images (Figs 1A 1B) theRSAP projection image displays a higher contrast andalso overcomes the influence of the choroidal vascu-lature on GA visualization as shown in Figure 5B
Figure 3 Intensity profiles of columns 100 (A) and 370 (B) marked with the dashed blue and yellow lines in Figure 2A respectively
Red curve intensity values of two columns Blue circle local maximum intensity point Magenta star maximum intensity point at higher
depths (in the x direction) (A) Intensity distribution in GA region (B) Intensity distribution in normal region
Figure 4 Flowchart of the proposed technique
6 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Evaluation Study
To evaluate the RSAP technique for GA visualiza-tion 99 3D SD-OCTmacular images from 27 eyes of 21patients presenting with advanced nonexudative AMDwere collected and analyzed The data was acquiredlongitudinally with several observations from the sameeye as a GA was followed up in regular clinicalpractice All eyes were previously diagnosed with GA
and collected consecutively from clinical practice The
study protocol was approved by an institutional review
board and the HIPAA compliant research adhered to
the Declaration of Helsinki and all federal and state
laws Each 3D OCT data set was acquired over a 63 6
mm area (corresponding to 512 3 128 pixels) with a
1024-pixel (2 mm) axial resolution on a commercial SD-
OCT device (Cirrus OCT Carl Zeiss Meditec Inc)
Figure 5 GA projection image with RSAP (A) RSAP projection image (B) Results by the three discussed techniques in regions of interest
as indicated in (A)
Figure 6 Depth of sub-RPE region for GA separability for one SD-OCT scan Four GA projection images corresponding to four red points
are inset
7 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Both qualitative analysis and quantitative evalua-tions were performed on the data For the quantita-tive evaluation we computed two metrics to measurethe GA contrast and distinction
Mean difference (MD) between GA and back-ground regions defined by
MD frac14lGA lbg
lGA thorn lbg
eth1THORN
where lGA and lbg represent the average intensity valuesin GA and background regions respectively The GAand background regions are outlined manually in GAprojection imagesAn examplemanual annotationof theGAboundaries is shown in the top-left image in Figure 6markedwithagreenoutlineWeused theMDtomeasureGA contrast against background regions Higher MDvalues corresponded to higher contrast levels
GA separability of the GA projection image wasdefined as
SGA frac14 1max254
ifrac141
XK
jfrac141
jBiethjTHORN GethjTHORNj
Keth2THORN
where Bi indicates the binarized result with a globalthreshold of value i with the GA projection imagenormalized to the interval [0 255] and G indicates aGA segmentation result outlined manually K denotesthe number of the pixel in the GA projection image InBi andG the GA and background regions are binarizedto 1 and 0 respectively The reasoning behind Equation2 is to calculate the maximum overlap ratio between allof the binarization results generated with the global
threshold and a lsquolsquogold standardrsquorsquo outlined in the SVPprojection images by an expert ophthalmic grader Therange of separability values is from 0 to 1 The best GAseparability (ie SGA frac14 1) represents the GA regionsthat can be completely separated from the backgroundby a global threshold namely the binarization resultsgenerated with the global threshold and the lsquolsquogoldstandardrsquorsquo are the same This metric was used tomeasure the GA distinction Since one purpose ofgenerating GA projection images is to reliably deter-mine GA extent28 this separability measurementreflects the feasibility in the generated images
Results
The proposed RSAP technique was tested andcompared with the SVP and Sub-RPE Slab tech-niques qualitatively and quantitatively The influenceof the parameter controlling the maximum considereddepth for the sub-RPE region mentioned earlier wasfirst evaluated and later fixed for all 99 test images
Parameter Evaluation
Figure 6 shows the GA separability for differentdepths of sub-RPE regions in one SD-OCT imagewhere the depth was varied from 100 to 300 pixels(approximately 02~06 mm in the collected images)with an interval of 10 pixels Figure 6 demonstrates thatthe GA separability increases when the depth increasesfrom 100 to 240 pixels and then remains stable withfurther depth increases Figure 7 shows the relationship
Figure 7 Mean (redline) and standard deviation (pale pink shading) of the GA separability across the 99 SD-OCT images in this study
considering different values of depth of sub-RPE region
8 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Figure 1 Choroidal vasculature influence on GA visualization (A) SVP projection image (B) Sub-RPE Slab projection image (C) ExampleB-scan (D) Example B-scan with structure labels (E) Sub-CSI Slab projection image
3 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
regions where GA is present and darker backgroundregions) of the Sub-RPE Slab projection image isbetter than that of the SVP projection image becausethe Sub-RPE Slab projection excludes the highlyreflective layers above RPE such as the retinal nervefiber layer (RNFL) and ellipsoid zone (EZ) whichmay influence visualization as they are added onto theprojection image Due to the presence of thechoroidal vasculature within the region affected byGA we can observe dark regions within the GAextent in both of the projection images as markedwith the orange circle in Figure 1A
In order to overcome the influence of the choroidalvasculature on GA visualization one could followtwo different strategies (1) excluding the choroidalvasculature with a new Sub- Choroidal-Sclera Inter-face (CSI) Slab technique similar to that done withthe Sub-RPE Slab technique but restricting theprojection region beneath the CSI boundary or (2)detecting and filling the dark regions where choroidalvessels are present with high intensity values With thefirst strategy the choroidal vasculature would becompletely excluded because no vessels appearbeneath the CSI boundary but the main character-istic of GA in SD-OCT imaging choroidal brighten-ing would also be omitted As less light penetratesthrough the choroid the brightening differencebetween the regions inside and outside GA wouldalso be less dramatic reducing contrast An exampleillustrating this effect is shown in Figure 1E where theimage produced by the Sub-CSI Slab projectiontechnique shows less contrast for GA visualizationthan the earlier proposed techniques (Figs 1A 1B)The CSI boundaries for Sub-CSI Slab projection weredelineated manually The Sub-RPE (Fig 1B) andSub-CSI (Fig 1E) projections are to generate the enface image by collapsing the pixels beneath the RPEand CSI respectively It does not mean pixel intensityis only plotted along the identified boundaries (RPEor CSI) The pixels in the narrow band beneath theRPE or CSI are collapsed to generate the projectionimages In this work we focus on the second proposedstrategy forming an image based in the projection ofintensity values restricted to the choroid region whilealso detecting and filling with high intensity values thedarker regions within the extent of GA that arecaused by vessel presence We call this technique therestricted summed-area projection (RSAP)
GA Visualization Based on RSAP
The main strategy of the RSAP technique lies inusing the intensity distribution beneath the RPE layer
as observed in SD-OCT images to fill the low intensityregions produced by the presence of choroidal vesselsVessel presence is identified by analyzing intensityprofiles An example of intensity distribution beneaththe RPE in a typical SD-OCT scan namely the regionbetween the two dashed green curves in Figure 1Cwhere GA is present is displayed in Figure 2 We canobserve that the intensity distribution decreasestoward the x direction (depth) and the rate of thisdecrease in the GA region is typically slower than thatin the normal region Figures 3A and 3B show twointensity profiles (marked with the red curve) of thesub-RPE region (Fig 2A) in the GA and the normalregions respectively The large concave regionpointed out in Figure 3A corresponds to the choroidalvessel marked with the orange oval in Figure 1D Thisvessel would have a great influence on GA visualiza-tion that is solely based on intensity summation suchas with the SVP and Sub-RPE Slab methods It can beseen from Figure 3 that for the same x position in thechoroidal region (axial location) the intensity valuesin the GA region (Fig 3A) are larger than those in thenormal region (Fig 3B) except in particular axiallocations (x axis) where choroidal vessel regions arelocated (marked with the orange oval circle in Fig 1Dand the dashed black circle in Fig 3A) In otherwords regions where GA is present will have higherintensity profiles than those where GA is not presentwhile vessel location can be identified with thepresence of local minimums in such profiles Thisprinciple constitutes the basic idea of the proposedRSAP technique
The flowchart of our technique is shown in Figure4 which comprises the following operations
(1) Segmentation of BM boundary the BM bound-aries were segmented with the automatic 3Dgraph search method27
(2) Flattening of sub-RPE region an image iscomposed by taking the recorded intensity valuesbeneath the segmented BM boundary up to amaximum depth where GA can be detectedconstituting a flattened sub-RPE image Thismaximum depth is set up as an independentparameter in this work (depth of sub-RPE asexplained in the later parameter evaluation) Theflattened sub-RPE region is shown in Figure 2A
(3) Finding local maximum intensity points For eachcolumn in the flattened sub-RPE region (A-scanlocation) the points with local maximum inten-sity value (namely the intensity value of the pointis larger than those of its two connected points)
4 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
are found as marked with the blue circle inFigure 3
(4) Locating maximum intensity points at higherdepths The maximum intensity points whosevalue follows a constantly decreasing functionwith depth (x axis) are selected as marked withthe magenta stars in Figure 3 The purpose of thisstep is to ensure a constantly descendent intensityprofile beneath RPE
(5) Calculating the area below the surface construct-ed with the maximum intensity points at higherdepths We interpolate the intensity profile in the
axial locations between the selected maximumintensity points at higher depths using linearinterpolation (magenta lines in Fig 3) andcalculate the area of the polygon formed by thisinterpolation and a baseline of zero intensity (areabelow the magenta lines marked in Fig 3)
(6) Taking the calculated area above as the primaryGA projection value at each projection locationand
(7) Using a median filter to smooth the generated GAprojection image to alleviate the noise influenceand make the final GA projection image smoother
Figure 2 Intensity distribution analysis (A) Flattened sub-RPE region (B) Intensity surface of (A)
5 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
a simple median filter with a 3 3 3 neighborhood
was used
Figure 5A shows the GA projection image with the
proposed RSAP technique and Figure 5B shows a
detail comparison of three GA projection techniques
in regions of interest corresponding to the red dashedrectangles in Figure 6A Compared with the SVP andSub-RPE Slab projection images (Figs 1A 1B) theRSAP projection image displays a higher contrast andalso overcomes the influence of the choroidal vascu-lature on GA visualization as shown in Figure 5B
Figure 3 Intensity profiles of columns 100 (A) and 370 (B) marked with the dashed blue and yellow lines in Figure 2A respectively
Red curve intensity values of two columns Blue circle local maximum intensity point Magenta star maximum intensity point at higher
depths (in the x direction) (A) Intensity distribution in GA region (B) Intensity distribution in normal region
Figure 4 Flowchart of the proposed technique
6 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Evaluation Study
To evaluate the RSAP technique for GA visualiza-tion 99 3D SD-OCTmacular images from 27 eyes of 21patients presenting with advanced nonexudative AMDwere collected and analyzed The data was acquiredlongitudinally with several observations from the sameeye as a GA was followed up in regular clinicalpractice All eyes were previously diagnosed with GA
and collected consecutively from clinical practice The
study protocol was approved by an institutional review
board and the HIPAA compliant research adhered to
the Declaration of Helsinki and all federal and state
laws Each 3D OCT data set was acquired over a 63 6
mm area (corresponding to 512 3 128 pixels) with a
1024-pixel (2 mm) axial resolution on a commercial SD-
OCT device (Cirrus OCT Carl Zeiss Meditec Inc)
Figure 5 GA projection image with RSAP (A) RSAP projection image (B) Results by the three discussed techniques in regions of interest
as indicated in (A)
Figure 6 Depth of sub-RPE region for GA separability for one SD-OCT scan Four GA projection images corresponding to four red points
are inset
7 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Both qualitative analysis and quantitative evalua-tions were performed on the data For the quantita-tive evaluation we computed two metrics to measurethe GA contrast and distinction
Mean difference (MD) between GA and back-ground regions defined by
MD frac14lGA lbg
lGA thorn lbg
eth1THORN
where lGA and lbg represent the average intensity valuesin GA and background regions respectively The GAand background regions are outlined manually in GAprojection imagesAn examplemanual annotationof theGAboundaries is shown in the top-left image in Figure 6markedwithagreenoutlineWeused theMDtomeasureGA contrast against background regions Higher MDvalues corresponded to higher contrast levels
GA separability of the GA projection image wasdefined as
SGA frac14 1max254
ifrac141
XK
jfrac141
jBiethjTHORN GethjTHORNj
Keth2THORN
where Bi indicates the binarized result with a globalthreshold of value i with the GA projection imagenormalized to the interval [0 255] and G indicates aGA segmentation result outlined manually K denotesthe number of the pixel in the GA projection image InBi andG the GA and background regions are binarizedto 1 and 0 respectively The reasoning behind Equation2 is to calculate the maximum overlap ratio between allof the binarization results generated with the global
threshold and a lsquolsquogold standardrsquorsquo outlined in the SVPprojection images by an expert ophthalmic grader Therange of separability values is from 0 to 1 The best GAseparability (ie SGA frac14 1) represents the GA regionsthat can be completely separated from the backgroundby a global threshold namely the binarization resultsgenerated with the global threshold and the lsquolsquogoldstandardrsquorsquo are the same This metric was used tomeasure the GA distinction Since one purpose ofgenerating GA projection images is to reliably deter-mine GA extent28 this separability measurementreflects the feasibility in the generated images
Results
The proposed RSAP technique was tested andcompared with the SVP and Sub-RPE Slab tech-niques qualitatively and quantitatively The influenceof the parameter controlling the maximum considereddepth for the sub-RPE region mentioned earlier wasfirst evaluated and later fixed for all 99 test images
Parameter Evaluation
Figure 6 shows the GA separability for differentdepths of sub-RPE regions in one SD-OCT imagewhere the depth was varied from 100 to 300 pixels(approximately 02~06 mm in the collected images)with an interval of 10 pixels Figure 6 demonstrates thatthe GA separability increases when the depth increasesfrom 100 to 240 pixels and then remains stable withfurther depth increases Figure 7 shows the relationship
Figure 7 Mean (redline) and standard deviation (pale pink shading) of the GA separability across the 99 SD-OCT images in this study
considering different values of depth of sub-RPE region
8 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
regions where GA is present and darker backgroundregions) of the Sub-RPE Slab projection image isbetter than that of the SVP projection image becausethe Sub-RPE Slab projection excludes the highlyreflective layers above RPE such as the retinal nervefiber layer (RNFL) and ellipsoid zone (EZ) whichmay influence visualization as they are added onto theprojection image Due to the presence of thechoroidal vasculature within the region affected byGA we can observe dark regions within the GAextent in both of the projection images as markedwith the orange circle in Figure 1A
In order to overcome the influence of the choroidalvasculature on GA visualization one could followtwo different strategies (1) excluding the choroidalvasculature with a new Sub- Choroidal-Sclera Inter-face (CSI) Slab technique similar to that done withthe Sub-RPE Slab technique but restricting theprojection region beneath the CSI boundary or (2)detecting and filling the dark regions where choroidalvessels are present with high intensity values With thefirst strategy the choroidal vasculature would becompletely excluded because no vessels appearbeneath the CSI boundary but the main character-istic of GA in SD-OCT imaging choroidal brighten-ing would also be omitted As less light penetratesthrough the choroid the brightening differencebetween the regions inside and outside GA wouldalso be less dramatic reducing contrast An exampleillustrating this effect is shown in Figure 1E where theimage produced by the Sub-CSI Slab projectiontechnique shows less contrast for GA visualizationthan the earlier proposed techniques (Figs 1A 1B)The CSI boundaries for Sub-CSI Slab projection weredelineated manually The Sub-RPE (Fig 1B) andSub-CSI (Fig 1E) projections are to generate the enface image by collapsing the pixels beneath the RPEand CSI respectively It does not mean pixel intensityis only plotted along the identified boundaries (RPEor CSI) The pixels in the narrow band beneath theRPE or CSI are collapsed to generate the projectionimages In this work we focus on the second proposedstrategy forming an image based in the projection ofintensity values restricted to the choroid region whilealso detecting and filling with high intensity values thedarker regions within the extent of GA that arecaused by vessel presence We call this technique therestricted summed-area projection (RSAP)
GA Visualization Based on RSAP
The main strategy of the RSAP technique lies inusing the intensity distribution beneath the RPE layer
as observed in SD-OCT images to fill the low intensityregions produced by the presence of choroidal vesselsVessel presence is identified by analyzing intensityprofiles An example of intensity distribution beneaththe RPE in a typical SD-OCT scan namely the regionbetween the two dashed green curves in Figure 1Cwhere GA is present is displayed in Figure 2 We canobserve that the intensity distribution decreasestoward the x direction (depth) and the rate of thisdecrease in the GA region is typically slower than thatin the normal region Figures 3A and 3B show twointensity profiles (marked with the red curve) of thesub-RPE region (Fig 2A) in the GA and the normalregions respectively The large concave regionpointed out in Figure 3A corresponds to the choroidalvessel marked with the orange oval in Figure 1D Thisvessel would have a great influence on GA visualiza-tion that is solely based on intensity summation suchas with the SVP and Sub-RPE Slab methods It can beseen from Figure 3 that for the same x position in thechoroidal region (axial location) the intensity valuesin the GA region (Fig 3A) are larger than those in thenormal region (Fig 3B) except in particular axiallocations (x axis) where choroidal vessel regions arelocated (marked with the orange oval circle in Fig 1Dand the dashed black circle in Fig 3A) In otherwords regions where GA is present will have higherintensity profiles than those where GA is not presentwhile vessel location can be identified with thepresence of local minimums in such profiles Thisprinciple constitutes the basic idea of the proposedRSAP technique
The flowchart of our technique is shown in Figure4 which comprises the following operations
(1) Segmentation of BM boundary the BM bound-aries were segmented with the automatic 3Dgraph search method27
(2) Flattening of sub-RPE region an image iscomposed by taking the recorded intensity valuesbeneath the segmented BM boundary up to amaximum depth where GA can be detectedconstituting a flattened sub-RPE image Thismaximum depth is set up as an independentparameter in this work (depth of sub-RPE asexplained in the later parameter evaluation) Theflattened sub-RPE region is shown in Figure 2A
(3) Finding local maximum intensity points For eachcolumn in the flattened sub-RPE region (A-scanlocation) the points with local maximum inten-sity value (namely the intensity value of the pointis larger than those of its two connected points)
4 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
are found as marked with the blue circle inFigure 3
(4) Locating maximum intensity points at higherdepths The maximum intensity points whosevalue follows a constantly decreasing functionwith depth (x axis) are selected as marked withthe magenta stars in Figure 3 The purpose of thisstep is to ensure a constantly descendent intensityprofile beneath RPE
(5) Calculating the area below the surface construct-ed with the maximum intensity points at higherdepths We interpolate the intensity profile in the
axial locations between the selected maximumintensity points at higher depths using linearinterpolation (magenta lines in Fig 3) andcalculate the area of the polygon formed by thisinterpolation and a baseline of zero intensity (areabelow the magenta lines marked in Fig 3)
(6) Taking the calculated area above as the primaryGA projection value at each projection locationand
(7) Using a median filter to smooth the generated GAprojection image to alleviate the noise influenceand make the final GA projection image smoother
Figure 2 Intensity distribution analysis (A) Flattened sub-RPE region (B) Intensity surface of (A)
5 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
a simple median filter with a 3 3 3 neighborhood
was used
Figure 5A shows the GA projection image with the
proposed RSAP technique and Figure 5B shows a
detail comparison of three GA projection techniques
in regions of interest corresponding to the red dashedrectangles in Figure 6A Compared with the SVP andSub-RPE Slab projection images (Figs 1A 1B) theRSAP projection image displays a higher contrast andalso overcomes the influence of the choroidal vascu-lature on GA visualization as shown in Figure 5B
Figure 3 Intensity profiles of columns 100 (A) and 370 (B) marked with the dashed blue and yellow lines in Figure 2A respectively
Red curve intensity values of two columns Blue circle local maximum intensity point Magenta star maximum intensity point at higher
depths (in the x direction) (A) Intensity distribution in GA region (B) Intensity distribution in normal region
Figure 4 Flowchart of the proposed technique
6 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Evaluation Study
To evaluate the RSAP technique for GA visualiza-tion 99 3D SD-OCTmacular images from 27 eyes of 21patients presenting with advanced nonexudative AMDwere collected and analyzed The data was acquiredlongitudinally with several observations from the sameeye as a GA was followed up in regular clinicalpractice All eyes were previously diagnosed with GA
and collected consecutively from clinical practice The
study protocol was approved by an institutional review
board and the HIPAA compliant research adhered to
the Declaration of Helsinki and all federal and state
laws Each 3D OCT data set was acquired over a 63 6
mm area (corresponding to 512 3 128 pixels) with a
1024-pixel (2 mm) axial resolution on a commercial SD-
OCT device (Cirrus OCT Carl Zeiss Meditec Inc)
Figure 5 GA projection image with RSAP (A) RSAP projection image (B) Results by the three discussed techniques in regions of interest
as indicated in (A)
Figure 6 Depth of sub-RPE region for GA separability for one SD-OCT scan Four GA projection images corresponding to four red points
are inset
7 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Both qualitative analysis and quantitative evalua-tions were performed on the data For the quantita-tive evaluation we computed two metrics to measurethe GA contrast and distinction
Mean difference (MD) between GA and back-ground regions defined by
MD frac14lGA lbg
lGA thorn lbg
eth1THORN
where lGA and lbg represent the average intensity valuesin GA and background regions respectively The GAand background regions are outlined manually in GAprojection imagesAn examplemanual annotationof theGAboundaries is shown in the top-left image in Figure 6markedwithagreenoutlineWeused theMDtomeasureGA contrast against background regions Higher MDvalues corresponded to higher contrast levels
GA separability of the GA projection image wasdefined as
SGA frac14 1max254
ifrac141
XK
jfrac141
jBiethjTHORN GethjTHORNj
Keth2THORN
where Bi indicates the binarized result with a globalthreshold of value i with the GA projection imagenormalized to the interval [0 255] and G indicates aGA segmentation result outlined manually K denotesthe number of the pixel in the GA projection image InBi andG the GA and background regions are binarizedto 1 and 0 respectively The reasoning behind Equation2 is to calculate the maximum overlap ratio between allof the binarization results generated with the global
threshold and a lsquolsquogold standardrsquorsquo outlined in the SVPprojection images by an expert ophthalmic grader Therange of separability values is from 0 to 1 The best GAseparability (ie SGA frac14 1) represents the GA regionsthat can be completely separated from the backgroundby a global threshold namely the binarization resultsgenerated with the global threshold and the lsquolsquogoldstandardrsquorsquo are the same This metric was used tomeasure the GA distinction Since one purpose ofgenerating GA projection images is to reliably deter-mine GA extent28 this separability measurementreflects the feasibility in the generated images
Results
The proposed RSAP technique was tested andcompared with the SVP and Sub-RPE Slab tech-niques qualitatively and quantitatively The influenceof the parameter controlling the maximum considereddepth for the sub-RPE region mentioned earlier wasfirst evaluated and later fixed for all 99 test images
Parameter Evaluation
Figure 6 shows the GA separability for differentdepths of sub-RPE regions in one SD-OCT imagewhere the depth was varied from 100 to 300 pixels(approximately 02~06 mm in the collected images)with an interval of 10 pixels Figure 6 demonstrates thatthe GA separability increases when the depth increasesfrom 100 to 240 pixels and then remains stable withfurther depth increases Figure 7 shows the relationship
Figure 7 Mean (redline) and standard deviation (pale pink shading) of the GA separability across the 99 SD-OCT images in this study
considering different values of depth of sub-RPE region
8 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
are found as marked with the blue circle inFigure 3
(4) Locating maximum intensity points at higherdepths The maximum intensity points whosevalue follows a constantly decreasing functionwith depth (x axis) are selected as marked withthe magenta stars in Figure 3 The purpose of thisstep is to ensure a constantly descendent intensityprofile beneath RPE
(5) Calculating the area below the surface construct-ed with the maximum intensity points at higherdepths We interpolate the intensity profile in the
axial locations between the selected maximumintensity points at higher depths using linearinterpolation (magenta lines in Fig 3) andcalculate the area of the polygon formed by thisinterpolation and a baseline of zero intensity (areabelow the magenta lines marked in Fig 3)
(6) Taking the calculated area above as the primaryGA projection value at each projection locationand
(7) Using a median filter to smooth the generated GAprojection image to alleviate the noise influenceand make the final GA projection image smoother
Figure 2 Intensity distribution analysis (A) Flattened sub-RPE region (B) Intensity surface of (A)
5 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
a simple median filter with a 3 3 3 neighborhood
was used
Figure 5A shows the GA projection image with the
proposed RSAP technique and Figure 5B shows a
detail comparison of three GA projection techniques
in regions of interest corresponding to the red dashedrectangles in Figure 6A Compared with the SVP andSub-RPE Slab projection images (Figs 1A 1B) theRSAP projection image displays a higher contrast andalso overcomes the influence of the choroidal vascu-lature on GA visualization as shown in Figure 5B
Figure 3 Intensity profiles of columns 100 (A) and 370 (B) marked with the dashed blue and yellow lines in Figure 2A respectively
Red curve intensity values of two columns Blue circle local maximum intensity point Magenta star maximum intensity point at higher
depths (in the x direction) (A) Intensity distribution in GA region (B) Intensity distribution in normal region
Figure 4 Flowchart of the proposed technique
6 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Evaluation Study
To evaluate the RSAP technique for GA visualiza-tion 99 3D SD-OCTmacular images from 27 eyes of 21patients presenting with advanced nonexudative AMDwere collected and analyzed The data was acquiredlongitudinally with several observations from the sameeye as a GA was followed up in regular clinicalpractice All eyes were previously diagnosed with GA
and collected consecutively from clinical practice The
study protocol was approved by an institutional review
board and the HIPAA compliant research adhered to
the Declaration of Helsinki and all federal and state
laws Each 3D OCT data set was acquired over a 63 6
mm area (corresponding to 512 3 128 pixels) with a
1024-pixel (2 mm) axial resolution on a commercial SD-
OCT device (Cirrus OCT Carl Zeiss Meditec Inc)
Figure 5 GA projection image with RSAP (A) RSAP projection image (B) Results by the three discussed techniques in regions of interest
as indicated in (A)
Figure 6 Depth of sub-RPE region for GA separability for one SD-OCT scan Four GA projection images corresponding to four red points
are inset
7 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Both qualitative analysis and quantitative evalua-tions were performed on the data For the quantita-tive evaluation we computed two metrics to measurethe GA contrast and distinction
Mean difference (MD) between GA and back-ground regions defined by
MD frac14lGA lbg
lGA thorn lbg
eth1THORN
where lGA and lbg represent the average intensity valuesin GA and background regions respectively The GAand background regions are outlined manually in GAprojection imagesAn examplemanual annotationof theGAboundaries is shown in the top-left image in Figure 6markedwithagreenoutlineWeused theMDtomeasureGA contrast against background regions Higher MDvalues corresponded to higher contrast levels
GA separability of the GA projection image wasdefined as
SGA frac14 1max254
ifrac141
XK
jfrac141
jBiethjTHORN GethjTHORNj
Keth2THORN
where Bi indicates the binarized result with a globalthreshold of value i with the GA projection imagenormalized to the interval [0 255] and G indicates aGA segmentation result outlined manually K denotesthe number of the pixel in the GA projection image InBi andG the GA and background regions are binarizedto 1 and 0 respectively The reasoning behind Equation2 is to calculate the maximum overlap ratio between allof the binarization results generated with the global
threshold and a lsquolsquogold standardrsquorsquo outlined in the SVPprojection images by an expert ophthalmic grader Therange of separability values is from 0 to 1 The best GAseparability (ie SGA frac14 1) represents the GA regionsthat can be completely separated from the backgroundby a global threshold namely the binarization resultsgenerated with the global threshold and the lsquolsquogoldstandardrsquorsquo are the same This metric was used tomeasure the GA distinction Since one purpose ofgenerating GA projection images is to reliably deter-mine GA extent28 this separability measurementreflects the feasibility in the generated images
Results
The proposed RSAP technique was tested andcompared with the SVP and Sub-RPE Slab tech-niques qualitatively and quantitatively The influenceof the parameter controlling the maximum considereddepth for the sub-RPE region mentioned earlier wasfirst evaluated and later fixed for all 99 test images
Parameter Evaluation
Figure 6 shows the GA separability for differentdepths of sub-RPE regions in one SD-OCT imagewhere the depth was varied from 100 to 300 pixels(approximately 02~06 mm in the collected images)with an interval of 10 pixels Figure 6 demonstrates thatthe GA separability increases when the depth increasesfrom 100 to 240 pixels and then remains stable withfurther depth increases Figure 7 shows the relationship
Figure 7 Mean (redline) and standard deviation (pale pink shading) of the GA separability across the 99 SD-OCT images in this study
considering different values of depth of sub-RPE region
8 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
a simple median filter with a 3 3 3 neighborhood
was used
Figure 5A shows the GA projection image with the
proposed RSAP technique and Figure 5B shows a
detail comparison of three GA projection techniques
in regions of interest corresponding to the red dashedrectangles in Figure 6A Compared with the SVP andSub-RPE Slab projection images (Figs 1A 1B) theRSAP projection image displays a higher contrast andalso overcomes the influence of the choroidal vascu-lature on GA visualization as shown in Figure 5B
Figure 3 Intensity profiles of columns 100 (A) and 370 (B) marked with the dashed blue and yellow lines in Figure 2A respectively
Red curve intensity values of two columns Blue circle local maximum intensity point Magenta star maximum intensity point at higher
depths (in the x direction) (A) Intensity distribution in GA region (B) Intensity distribution in normal region
Figure 4 Flowchart of the proposed technique
6 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Evaluation Study
To evaluate the RSAP technique for GA visualiza-tion 99 3D SD-OCTmacular images from 27 eyes of 21patients presenting with advanced nonexudative AMDwere collected and analyzed The data was acquiredlongitudinally with several observations from the sameeye as a GA was followed up in regular clinicalpractice All eyes were previously diagnosed with GA
and collected consecutively from clinical practice The
study protocol was approved by an institutional review
board and the HIPAA compliant research adhered to
the Declaration of Helsinki and all federal and state
laws Each 3D OCT data set was acquired over a 63 6
mm area (corresponding to 512 3 128 pixels) with a
1024-pixel (2 mm) axial resolution on a commercial SD-
OCT device (Cirrus OCT Carl Zeiss Meditec Inc)
Figure 5 GA projection image with RSAP (A) RSAP projection image (B) Results by the three discussed techniques in regions of interest
as indicated in (A)
Figure 6 Depth of sub-RPE region for GA separability for one SD-OCT scan Four GA projection images corresponding to four red points
are inset
7 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Both qualitative analysis and quantitative evalua-tions were performed on the data For the quantita-tive evaluation we computed two metrics to measurethe GA contrast and distinction
Mean difference (MD) between GA and back-ground regions defined by
MD frac14lGA lbg
lGA thorn lbg
eth1THORN
where lGA and lbg represent the average intensity valuesin GA and background regions respectively The GAand background regions are outlined manually in GAprojection imagesAn examplemanual annotationof theGAboundaries is shown in the top-left image in Figure 6markedwithagreenoutlineWeused theMDtomeasureGA contrast against background regions Higher MDvalues corresponded to higher contrast levels
GA separability of the GA projection image wasdefined as
SGA frac14 1max254
ifrac141
XK
jfrac141
jBiethjTHORN GethjTHORNj
Keth2THORN
where Bi indicates the binarized result with a globalthreshold of value i with the GA projection imagenormalized to the interval [0 255] and G indicates aGA segmentation result outlined manually K denotesthe number of the pixel in the GA projection image InBi andG the GA and background regions are binarizedto 1 and 0 respectively The reasoning behind Equation2 is to calculate the maximum overlap ratio between allof the binarization results generated with the global
threshold and a lsquolsquogold standardrsquorsquo outlined in the SVPprojection images by an expert ophthalmic grader Therange of separability values is from 0 to 1 The best GAseparability (ie SGA frac14 1) represents the GA regionsthat can be completely separated from the backgroundby a global threshold namely the binarization resultsgenerated with the global threshold and the lsquolsquogoldstandardrsquorsquo are the same This metric was used tomeasure the GA distinction Since one purpose ofgenerating GA projection images is to reliably deter-mine GA extent28 this separability measurementreflects the feasibility in the generated images
Results
The proposed RSAP technique was tested andcompared with the SVP and Sub-RPE Slab tech-niques qualitatively and quantitatively The influenceof the parameter controlling the maximum considereddepth for the sub-RPE region mentioned earlier wasfirst evaluated and later fixed for all 99 test images
Parameter Evaluation
Figure 6 shows the GA separability for differentdepths of sub-RPE regions in one SD-OCT imagewhere the depth was varied from 100 to 300 pixels(approximately 02~06 mm in the collected images)with an interval of 10 pixels Figure 6 demonstrates thatthe GA separability increases when the depth increasesfrom 100 to 240 pixels and then remains stable withfurther depth increases Figure 7 shows the relationship
Figure 7 Mean (redline) and standard deviation (pale pink shading) of the GA separability across the 99 SD-OCT images in this study
considering different values of depth of sub-RPE region
8 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Evaluation Study
To evaluate the RSAP technique for GA visualiza-tion 99 3D SD-OCTmacular images from 27 eyes of 21patients presenting with advanced nonexudative AMDwere collected and analyzed The data was acquiredlongitudinally with several observations from the sameeye as a GA was followed up in regular clinicalpractice All eyes were previously diagnosed with GA
and collected consecutively from clinical practice The
study protocol was approved by an institutional review
board and the HIPAA compliant research adhered to
the Declaration of Helsinki and all federal and state
laws Each 3D OCT data set was acquired over a 63 6
mm area (corresponding to 512 3 128 pixels) with a
1024-pixel (2 mm) axial resolution on a commercial SD-
OCT device (Cirrus OCT Carl Zeiss Meditec Inc)
Figure 5 GA projection image with RSAP (A) RSAP projection image (B) Results by the three discussed techniques in regions of interest
as indicated in (A)
Figure 6 Depth of sub-RPE region for GA separability for one SD-OCT scan Four GA projection images corresponding to four red points
are inset
7 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Both qualitative analysis and quantitative evalua-tions were performed on the data For the quantita-tive evaluation we computed two metrics to measurethe GA contrast and distinction
Mean difference (MD) between GA and back-ground regions defined by
MD frac14lGA lbg
lGA thorn lbg
eth1THORN
where lGA and lbg represent the average intensity valuesin GA and background regions respectively The GAand background regions are outlined manually in GAprojection imagesAn examplemanual annotationof theGAboundaries is shown in the top-left image in Figure 6markedwithagreenoutlineWeused theMDtomeasureGA contrast against background regions Higher MDvalues corresponded to higher contrast levels
GA separability of the GA projection image wasdefined as
SGA frac14 1max254
ifrac141
XK
jfrac141
jBiethjTHORN GethjTHORNj
Keth2THORN
where Bi indicates the binarized result with a globalthreshold of value i with the GA projection imagenormalized to the interval [0 255] and G indicates aGA segmentation result outlined manually K denotesthe number of the pixel in the GA projection image InBi andG the GA and background regions are binarizedto 1 and 0 respectively The reasoning behind Equation2 is to calculate the maximum overlap ratio between allof the binarization results generated with the global
threshold and a lsquolsquogold standardrsquorsquo outlined in the SVPprojection images by an expert ophthalmic grader Therange of separability values is from 0 to 1 The best GAseparability (ie SGA frac14 1) represents the GA regionsthat can be completely separated from the backgroundby a global threshold namely the binarization resultsgenerated with the global threshold and the lsquolsquogoldstandardrsquorsquo are the same This metric was used tomeasure the GA distinction Since one purpose ofgenerating GA projection images is to reliably deter-mine GA extent28 this separability measurementreflects the feasibility in the generated images
Results
The proposed RSAP technique was tested andcompared with the SVP and Sub-RPE Slab tech-niques qualitatively and quantitatively The influenceof the parameter controlling the maximum considereddepth for the sub-RPE region mentioned earlier wasfirst evaluated and later fixed for all 99 test images
Parameter Evaluation
Figure 6 shows the GA separability for differentdepths of sub-RPE regions in one SD-OCT imagewhere the depth was varied from 100 to 300 pixels(approximately 02~06 mm in the collected images)with an interval of 10 pixels Figure 6 demonstrates thatthe GA separability increases when the depth increasesfrom 100 to 240 pixels and then remains stable withfurther depth increases Figure 7 shows the relationship
Figure 7 Mean (redline) and standard deviation (pale pink shading) of the GA separability across the 99 SD-OCT images in this study
considering different values of depth of sub-RPE region
8 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Both qualitative analysis and quantitative evalua-tions were performed on the data For the quantita-tive evaluation we computed two metrics to measurethe GA contrast and distinction
Mean difference (MD) between GA and back-ground regions defined by
MD frac14lGA lbg
lGA thorn lbg
eth1THORN
where lGA and lbg represent the average intensity valuesin GA and background regions respectively The GAand background regions are outlined manually in GAprojection imagesAn examplemanual annotationof theGAboundaries is shown in the top-left image in Figure 6markedwithagreenoutlineWeused theMDtomeasureGA contrast against background regions Higher MDvalues corresponded to higher contrast levels
GA separability of the GA projection image wasdefined as
SGA frac14 1max254
ifrac141
XK
jfrac141
jBiethjTHORN GethjTHORNj
Keth2THORN
where Bi indicates the binarized result with a globalthreshold of value i with the GA projection imagenormalized to the interval [0 255] and G indicates aGA segmentation result outlined manually K denotesthe number of the pixel in the GA projection image InBi andG the GA and background regions are binarizedto 1 and 0 respectively The reasoning behind Equation2 is to calculate the maximum overlap ratio between allof the binarization results generated with the global
threshold and a lsquolsquogold standardrsquorsquo outlined in the SVPprojection images by an expert ophthalmic grader Therange of separability values is from 0 to 1 The best GAseparability (ie SGA frac14 1) represents the GA regionsthat can be completely separated from the backgroundby a global threshold namely the binarization resultsgenerated with the global threshold and the lsquolsquogoldstandardrsquorsquo are the same This metric was used tomeasure the GA distinction Since one purpose ofgenerating GA projection images is to reliably deter-mine GA extent28 this separability measurementreflects the feasibility in the generated images
Results
The proposed RSAP technique was tested andcompared with the SVP and Sub-RPE Slab tech-niques qualitatively and quantitatively The influenceof the parameter controlling the maximum considereddepth for the sub-RPE region mentioned earlier wasfirst evaluated and later fixed for all 99 test images
Parameter Evaluation
Figure 6 shows the GA separability for differentdepths of sub-RPE regions in one SD-OCT imagewhere the depth was varied from 100 to 300 pixels(approximately 02~06 mm in the collected images)with an interval of 10 pixels Figure 6 demonstrates thatthe GA separability increases when the depth increasesfrom 100 to 240 pixels and then remains stable withfurther depth increases Figure 7 shows the relationship
Figure 7 Mean (redline) and standard deviation (pale pink shading) of the GA separability across the 99 SD-OCT images in this study
considering different values of depth of sub-RPE region
8 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
between the mean and standard deviation of the GAseparability and the depth of sub-RPE region for all 99SD-OCT images which indicates an optimal depthrange from 190 to 210 pixels (approximately 037~041mm) When the depth of sub-RPE region is muchsmaller than the axial region where GA can be observedin the SD-OCT images the GA separability is lowbecause only a limited region of the high-intensityvalues associated with GA near the BM are used Whenthis depth is increased to values that are larger thanoptimal the GA separability decreases because morebackground intensity in sclera is included In this paperthe depth of sub-RPE region was set to be 200 pixels(~039 mm) for all test images
Quantitative Results and Qualitative Analysis
Table 1 shows the average performance of thethree tested techniques where all of 99 cubes (3D SD-OCT images) from 27 eyes in 21 patients are usedFigure 8 shows the MD and separability values foreach case included in our analysis
Figures 9A to C show the GA projection imagesgenerated from the SD-OCT scan of a patientrsquos left eye
using the three techniques tested SVP Sub-RPE Slaband RSAP respectively Figure 9D shows a one-lineprofile of the rows of Figures 9A to C marked with adashed yellow line The region marked with the dashedblack oval in Figure 9D corresponds to the brightregions in the GA projection images (Figs 9AndashC)
Another example comparison of the SVP Sub-RPE Slab and RSAP projection images is shown inFigures 10A to 10C respectively Figure 10D showsthe results in the region of interest marked with thered dashed rectangle in Figure 10A and Figure 10Eshows one B-scan corresponding to the row in the GAprojection images marked with a dashed yellow line
Discussion
We have presented a novel technique the RSAPto increase the contrast and distinction of GA in afundus projection image In addition we comparedthe images produced with the ones produced by twoknown methods the SVP and Sub-RPE Slab tech-niques Identifying and quantifying GA area isbecoming more important in the diagnosis andmanagement of advanced dry AMD11 With thedevelopment of pharmacologic and cell-based thera-pies for GA accurately identifying and monitoringGA over time will be important in order to clinicallydetermine the efficacy or failures of these noveltreatment modalities2930 Although SD-OCT has thepotential to become the preferable technique forimaging the retina direct visualization of GA in SD-OCT has been limited by low contrast and overlap-ping of retinal pathologies and structures when
Table 1 Average MD and SGA of the SVP Sub-RPESlab and RSAP Projection Images
MethodsCompared
Number ofEyesCubes MD SGA
SVP 2799 0129 088Sub-RPE Slab 2799 0238 0919RSAP 2799 0276 0938
Figure 8 Comparison of the three tested GA projection techniques (A) Mean difference (B) GA separability
9 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
generating a fundus image The RSAP methodpresented here improves direct visualization of GAby considering a restricted projection and thecontribution of the choroidal vasculature
We did not adopt the overlap ratio of the numberof visualized lesions proposed in other studies 1622 toevaluate performance across techniques as GApresence can be easily identified with each techniquebut rather we focused on the contrast and distinctionbetween the GA and background regions amongdifferent techniques Based on these proposed twometrics the SVP technique had the lowest perfor-mance in the cases evaluated while our technique(RSAP) had the highest performance (Table 1)
GA separability with the RSAP technique washigher than both the SVP and Sub-RPE Slabtechniques and the RSAP technique produced imageswith a higher MD between GA and backgroundregions for most of cases (Fig 8) For several cases the
MD values of SVP are lower than 0 (Fig 8A) implyingthat in these SVP projection images the averageintensity values in GA regions are lower than that inbackground regions For most cases (9099frac14 91) theGA separability of the RSAP projection images ishigher than 09 (Fig 8B) suggesting a relatively highGA segmentation precision in RSAP projection imagesif an optimum global threshold can be obtained
Figure 9 displays an example where the GAcontrast in the SVP projection image (Fig 9A) is toolow to be able to easily distinguish the GA regions TheSub-RPE Slab (Fig 9B) and RSAP (Fig 9C)projection images have a relatively good contrast forGA visualization The RSAP projection image has thebest contrast because the background intensity nearthe GA boundary is the lowest (the green line is belowthe blue and red lines as shown in Fig 9D) Theintensity gradient of the GA boundary is higher in theRSAP projection image than that in the SVP and Sub-
Figure 9 Comparison of three GA projection techniques in the left eye of one patient (A) SVP (B) Sub-RPE Slab (C) RSAP (D) One-line
profile
10 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
RPE Slab projection images which would be helpfulfor enabling automated GA segmentation algorithms
From Figure 10 we can observe that the GAcontrast of the Sub-RPE Slab and RSAP projectionimages is better than that of the SVP projection imagebecause these two region-restricted techniques excludehighly reflective retinal layers above the BM Theintensity throughout the GA regions in the RSAPprojection image also seems more consistent than inboth the SVP and Sub-RPE Slab projection imagesbecause the choroidal vessel influence (marked withthree dashed yellow arrows in Fig 10E) has beenexcluded in our technique However the RSAPtechnique still presents some limitations due to thelower intensity stripes in the choroidal and scleralregions (marked with the dashed green arrow in Fig10E) there are still dark areas within GA regions asmarked with the dashed green arrow in Figure 10COn the other hand this limitation also exists in the
SVP and Sub-RPE Slab projection images It appears
that the dark stripes correspond to the presence of
retinal vessels suggesting that one solution may be to
detect and exclude the retinal vessels before generat-
ing GA projection images in a similar manner as that
which is done for the choroid vessels in this work
This will be a matter for future work
In conclusion we present a projection technique
from 3D SD-OCT images based on intensity distri-
bution in sub-RPE regions for the visualization of
GA which we called the RSAP technique The RSAP
technique improves on the previous methods by
considering and using the intensity distribution
characteristics in sub-RPE regions Quantitative
comparison in 99 3D SD-OCT scans from 21 patients
demonstrated that the RSAP is more effective for GA
visualization than the SVP and Sub-RPE Slab due to
increased GA contrast and distinction
Figure 10 Qualitative comparison of GA visualization in a patientrsquos left eye (A) SVP (B) Sub-RPE Slab (C) RSAP (D) Detail in region of
interest for the three tested techniques (E) Example B-scan at the location marked by the dashed yellow line in (AndashC)
11 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
Acknowledgments
Supported by a grant from the FundamentalResearch Funds for the Central Universities(30920140111004) a six talent peaks project inJiangsu Province (2014-SWYY-024) the Qing LanProject the Bio-X Interdisciplinary Initiatives Pro-gram of Stanford University and a Spectrum-SPADA innovation grant from Stanford University(NIH-NCATS-CTSA-SPECTRUM)
References
1 Bird AC Bressler NM Bressler SB et al Aninternational classification and grading systemfor age-related maculopathy and age-relatedmacular degeneration The International ARMEpidemiology Study Group Surv Ophthalmol199539367ndash374
2 Resnikoff S Pascolini D Etyarsquoals D et al Globaldata on visual impairment in the year 2002 BullWorld Health Organ 200482844ndash851
3 Bressler NM Bressler SB Fine SL Age-relatedmacular degeneration Surv Ophthalmol 198832375ndash413
4 Sunness JS Gonzalez-Baron J Applegate CA etal Enlargement of atrophy and visual acuity lossin the geographic atrophy form of age-relatedmacular degeneration Ophthalmology 19991061768ndash1779
5 Yehoshua Z Rosenfeld PJ Albini TA Currentclinical trials in dry AMD and the definition ofappropriate clinical outcome measures SeminOphthalmol 201126167ndash180
6 Bearelly S Chau FY Koreishi A et al Spectraldomain optical coherence tomography imaging ofgeographic atrophy margins Ophthalmol 20091161762ndash1769
7 Dunaief JL Dentchev T Ying GS Milam AHThe role of apoptosis in age-related maculardegeneration Arch Ophthalmol 20021201435ndash1442
8 Lutty G Grunwald J Majji AB Uyama MYoneya S Changes in choriocapillaris and retinalpigment epithelium in age-related macular degen-eration Mol Vis 1999535
9 McLeod DS Taomoto M Otsuji T et alQuantifying changes in RPE and choroidalvasculature in eyes with age-related macular
degeneration Invest Ophthalmol Vis Sci 2002431986ndash1993
10 Gregori G Wang F Rosenfeld PJ et al Spectraldomain optical coherence tomography imaging ofdrusen in nonexudative age-related maculardegeneration Ophthalmol 20111181373ndash1379
11 Yehoshua Z Rosenfeld PJ Grgori G et alProgression of geographic atrophy in age relatedmacular degeneration imaged with spectral do-main optical coherence tomography Ophthalmol-ogy 2011118679ndash686
12 Fleckenstein M Schmitz-Valckenberg S MartensC et al Fundus autofluorescence and spectral-domain optical coherence tomography character-istics in a rapidly progressing form of geographicatrophy Invest Ophthalmol Vis Sci 2011523761ndash3766
13 Fleckenstein M Schmitz-Valckenberg S AdrionC et al Tracking progression with spectral-domain optical coherence tomography in geo-graphic atrophy caused by age-related maculardegeneration Invest Ophthalmol Vis Sci 2010513846ndash3852
14 Gobel AP Fleckenstein M Schmitz-ValckenbergS Brinkmann CK Holz FG Imaging geographicatrophy in age-related macular degenerationOphthalmologica 2011226182ndash190
15 Bindewald A Bird AC Dandekar SS et alClassification of fundus autofluorescence patternsin early age-related macular disease InvestOphthalmol Vis Sci 2005463309ndash3314
16 Schmitz-Valckenberg S Fleckenstein M GobelAP Hohman TC Holz FG Optical coherencetomography and autofluorescence findings inareas with geographic atrophy due to age-relatedmacular degeneration Invest Ophthalmol Vis Sci2011521ndash6
17 Sayegh RG Simader C Scheschy U et al Asystematic comparison of spectral-domain opticalcoherence tomography and fundus autofluores-cence in patients with geographic atrophyOphthalmology 20111181844ndash1851
18 Chen Q Leng T Zheng LL Kutzscher L DeSisternes L Rubin DL An improved opticalcoherence tomography-derived fundus projectionimage for drusen visualization Retina 201434996ndash1005
19 Jiao S Knighton R Huang X Gregori GPuliafito C Simultaneous acquisition of sectionaland fundus ophthalmic images with spectral-domain optical coherence tomography OptExpress 200513444ndash452
12 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
20 Khanifar AA Koreishi AF Izatt Ja Toth CADrusen ultrastructure imaging with spectraldomain optical coherence tomography in age-related macular degeneration Ophthalmol 20081151883ndash1890
21 Wojtkowski M Srinivasan V Fujimoto JG et alThree-dimensional retinal imaging with high-speed ultrahigh-resolution optical coherence to-mography Ophthalmol 20051121734ndash1746
22 Stopa M Bower BA Davies E et al Correlationof pathologica features in spectral domain opticalcoherence tomography with conventional retinalstudies Retina 200828298ndash308
23 Yehoshua Z Garcia Filho CA Penha FM et alComparison of geographic atrophy measure-ments from the OCT fundus image and the sub-RPE slab image Ophthalmic Surg Lasers ImagingRetina 201344127ndash132
24 Chen Q Leng T Niu S et al A false color fusionstrategy for drusen and GA visualization in OCTimages Retina 2014342346ndash2358
25 Stetson PF Yehoshua Z Garcia Filho CA et alOCT minimum intensity as a predictor of
geographic atrophy enlargement Invest Ophthal-mol Vis Sci 201455792ndash800
26 Hogan MJ Alvarado JA Wddell JE Histologyof the human eye Philadelphia WB SaundersCo 1971
27 Garvin MK Abramoff MD Wu X Russell SRBurns TL Sonka M Automated 3-D intraretinallayer segmentation of macular spectral-domainoptical coherence tomography images IEEETrans Med Imaging 2009281436ndash1447
28 Chen Q De Sisternes L Leng T Zheng LLKutzscher L Rubin DL Semi-automatic geo-graphic atrophy segmentation for SD-OCT im-ages Biomed Optics Exp 201342729ndash2750
29 Volz C Pauly D Antibody therapies and theirchallenges in the treatment of age-related maculardegeneration Eur J Pharm Biopharm 2015piiS0939-6411(15)00102-2
30 Zhang K Hopkins JJ Heier JS et al Ciliaryneurotrophic factor delivered by encapsulated cellintraocular implants for treatment of geographicatrophy in age-related macular degenerationProc Natl Acad Sci U S A 20111086241ndash6245
13 TVST j 2015 j Vol 4 j No 5 j Article 2
Chen et al
