Excitation-and-collection geometry insensitive fluorescence imaging

aflp

Excitation-and-collection geometry insensitivefluorescence imaging of tissue-simulating turbid media

Jianan Y. Qu, Zhijian Huang, and Jianwen Hua

We present an imaging technique for the correction of geometrical effects in fluorescence measurementof optically thick, turbid media such as human tissue. Specifically, we use the cross-polarization methodto reject specular reflection and enhance the diffusive backscattering of polarized fluorescence excitationlight from the turbid media. We correct the nonuniformity of the image field caused by the excitation-and-collection geometry of a fluorescence imaging system by normalizing the fluorescence image to thecross-polarized reflection image. The ratio image provides a map of relative fluorescence yield, definedas the ratio of emerging fluorescence power to incident excitation, over the surface of an imaged homo-geneous turbid medium when fluorescence excitation-and-collection geometries vary in a wide range.We investigate the mechanism of ratio imaging by using Monte Carlo modeling. Our findings show thatthis technique could have a potential use in the detection of early cancer, which usually starts from asuperficial layer of tissue, based on the contrast in the tissue fluorescence of an early lesion and of thesurrounding normal tissue. © 2000 Optical Society of America

OCIS codes: 170.0170, 110.7050, 170.3880, 170.4580, 170.6510.

1. Introduction

The light-induced fluorescence ~LIF! imaging tech-nique has been extensively investigated as an effec-tive and noninvasive method with which to diagnosediseased tissue, especially early stage cancers.1–12

Tissue autofluorescence is emitted from endogenousfluorophores excited by low-power, short-wavelengthlight. The autofluorescence measured on the tissuesurface reveals some of the biochemical and biomor-phological changes that are occurring in the tissue.Careful in vivo studies of several organ sites by cal-ibrated fluorescence spectroscopy has shown that thetissue autofluorescence of early lesions is almost al-ways less than that of the surrounding normal tis-sues, although the spectral line shapes of earlylesions and normal tissues vary both from individualto individual and within an individual.5,7,8 Tissueutofluorescence is not only a function of endogenousuorophores but is affected by the tissue’s opticalroperties as well, because tissue is a kind of turbid

The authors are with the Department of Electrical and Elec-tronic Engineering, Hong Kong University of Science and Technol-ogy, Clear Water Bay, Kowloon, Hong Kong, China. J. Y. Qu’se-mail address is [email protected].

Received 7 June 1999; revised manuscript received 16 February2000.

0003-6935y00y193344-13$15.00y0© 2000 Optical Society of America

3344 APPLIED OPTICS y Vol. 39, No. 19 y 1 July 2000

medium. Strong evidence indicates that a robustalgorithm for discriminating a lesion from surround-ing normal tissue should include information on thetissue’s autofluorescence.

The calibrated fluorescence spectroscopy for in vivostudy of tissue autofluorescence commonly utilizes amultiple optical fiber sensor to deliver the excitationlight and collect the backscattered fluorescence sig-nal.5,7,8,12 The distal tip of the sensor is gentlytouched to the tissue surface to ensure constant flu-orescence excitation-and-collection geometry for mea-surements at different sites. This point-by-pointdiagnosis method has been successfully used for de-tection of early cancers at various organ sites.7,8

However, it is time consuming and not practical inclinical practice to examine a large area of tissue. Itwould be desirable to determine an imaging tech-nique. For examination of a large area by an imag-ing device the recorded fluorescence is stronglyaffected by geometrical effects such as separation ofthe source–sample–detector, incident–emission an-gles, and irregularity of the sample surface. With-out a mechanism to compensate for such geometricaleffects, it is impossible to map the fluorescence in theimaged area, especially for in vivo measurements.

Several groups of researchers have studied geomet-rical effects on measurement of tissue fluorescence andexplored various methods for determination of the rel-ative fluorescence on tissue surfaces. Profio et al.2and Lenz4 used a nonimaging approach to compensate

tm

lcpeiasafltodIs

for the inhomogeneity of fluorescence excitation andcollection in tissue. In their studies the fluorescencesignals were normalized to reflection signals of excita-tion light collected from the exactly same sites. Theresults showed that the geometrical effects could becorrected but that the artifacts caused by specular re-flection from the tissue surface created many false pos-itives. Analytical relations between fluorescence anddiffuse reflectance have been developed,13,14 based onphoton migration and diffusion theories. It has beenproved that, with knowledge of the tissue’s opticalproperties, the diffuse reflectance at the wavelengthsof fluorescence emission can be used to correct thedistortion of the autofluorescence caused by the scat-tering and absorption of tissue. Further, Wu et al.stated that the ratio of fluorescence to effective reflec-tance at the same wavelength could correct the geo-metrical effects in fluorescence measurement.13

Moreover, they pointed out that specular reflectionfrom a tissue surface must be carefully avoided or sub-tracted in the clinical application of the ratio method.Assuming that the scattering properties vary little indifferent samples of the same tissue, they demon-strated that the ratio technique can be directly used toextract a tissue’s intrinsic fluorescence and correct thegeometrical effects. In the research reported in Ref. 9the calibrated fluorescence spectral images were ob-tained from the tissue surface in a special geometry byuse of a combined ultrasonic and fluorescence spectro-scopic system. The technique may be invalid whenthe angular distribution of the backscattered ultra-sonic signal is different from that of the fluorescencesignal. Wang et al. reported an image-processingmethod for measurement of relative fluorescence onthe tissue surface.10 In a fully digital approach the rawfluorescence image of tissue is normalized to a refer-ence image, which is a processed raw fluorescence im-age by use of the moving-average algorithm, a type oflow-pass filter. To ensure that the lesions are filteredout and the excitation and collection nonuniformitiesare retained in the reference image, the algorithm re-quires that the lesion width be much smaller than theimaged area. The variation of fluorescence was as-sumed to be a slowly varying function over the tissuesurface, because a low-pass spatial filter cannot distin-guish a lesion from a geometrical artifact whose spatialfrequency is distributed in the same region as the le-sion.

In this paper we present a ratio imaging methodthat maps the relative fluorescence on the surface of ahomogeneous turbid medium with optical properties inthe range of human tissue. The technique utilizes theratio of fluorescence ~F! to diffuse reflectance ~R! of theexcitation light collected by a LIF imaging system tocorrect the geometrical effects in the fluorescence mea-surement. Our study was motivated by the well-established ratio technique reported in Ref. 13 and theresearch of Profio2 and Lenz.4 We analyze the tech-nique by using the method of Monte Carlo modeling.The technique is then evaluated experimentally with aseries of homogeneous and inhomogeneous tissuephantoms in various excitation-and-collection geome-

tries. We demonstrate that the technique could im-age an early lesion based on the contrast inautofluorescence between the lesion and normal tis-sue. Finally, qualitative discussions on modeling andexperimental results are presented.

2. Methods

For a conventional LIF imaging system the fluores-cence and reflectance of excitation light are easilymeasurable. The reflected light from the turbid me-dium consists of three components: specular reflec-tion, nearly diffusive backscattering ~scattered 1–2times!, and diffusive backscattering ~specular reflec-ion does not carry any information about nonunifor-ity of illumination and collection!. Wu et al. have

proved that the analytical relationships between flu-orescence and diffuse reflectance measured at thesame wavelength can be used to correct geometricaleffects in fluorescence measurement.13 The ques-tion is whether one can use the diffuse reflectance ofexcitation light, one of the easily measured constitu-ents of a LIF imaging system to correct geometricaleffects in fluorescence measurement.

To find the answer to the question we first inves-tigated the characteristics of fluorescence and diffusereflectance of excitation light on a tissue surface. In-stead of taking the approaches based on diffusion orphoton migration theory, we chose the method ofMonte Carlo modeling with which to calculate thetissue fluorescence and reflectance. Monte Carlomodeling does not assume that fluorescence and re-flectance are diffusive waves, as do diffusion and pho-ton migration theories. However, the majordisadvantage of Monte Carlo modeling is that it istime consuming to generate data with a reasonablygood signal-to-noise ratio ~SNR!. Briefly, one calcu-ates the distribution of excitation light in tissue byonvoluting that of an infinitely narrow beam and therofile of the illumination beam. One calculates thescape function of the endogenous fluorophore at var-ous depths by treating the emission of fluorophoress an isotropic radiating point source inside the tis-ue. The product of the distribution of excitationnd the fluorescence excitation efficiency yields theuorescence source distribution in tissue. Finally,he fluorescence emerging from the tissue surface isbtained by convolution of the fluorescence sourceistribution and the fluorescence escape function.n general, the fluorescence collected from the tissueurface can be calculated by

CF~r, lem! 5 *0

`

FF~r9, z, lex, lem!

# Gescape~r9, z, lem!dz, (1)

FF~r, z, lex, lem! 5 FEx~r, z, lex!

3 b~r, z, lem!, (2)

Fex~r, z, lex! 5 FEx9~r9, z, lex!

# Iprofile~r9, lex!, (3)

1 July 2000 y Vol. 39, No. 19 y APPLIED OPTICS 3345

flc

ecFdaVteR

~tTFiM

cf

c

3

where FF~r, z, lex, lem! and Gescape~r, z, lem! are theuorescence source distribution and fluorescence es-ape functions in the tissue, respectively. FEx~r, z,

lex! and b~r, z, lem! in Eq. ~2! are the distribution ofxcitation light and the fluorescence excitation effi-iency at the emission wavelength, respectively.Ex9~r, z, lex! and Iprofile~r, lex! in Eq. ~3! are theistribution of an infinitely narrow excitation beamnd the true illumination beam profile, respectively.

refers to the convolution of two functions. A de-ailed description and discussion of Monte Carlo mod-ling for fluorescence and reflectance can be found inefs. 8 and 15.In the simulation, collimated uniform excitation

20 mm in diameter! and a semi-infinite tissuelikeurbid medium with a flat surface were assumed.he illumination-and-collection geometry is shown inig. 1~a!, in which u and f are the collection and

llumination angles, respectively. We modified theonte Carlo code of Wang and Jacques15 to enable it

to generate angularly and radially resolved fluores-cence and reflectance at various illumination angles

Fig. 1. Monte Carlo modeling of fluorescence and diffuse reflec-tance of excitation light. The optical properties of the tissue at theexcitation wavelength are ma 5 16.5 cm21, ms 5 184 cm21, and g 50.9 and at fluorescence wavelength are ma 5 8.0 cm21, ms 5 145m21, and g 5 0.87. ~a! Illumination-and-collection geometry:[~0°, 90°! and u[~290°, 90°!. ~b! Normalized angular distribu-

tion of fluorescence and diffuse reflectance emerging from the tis-sue surface. Solid circles, illumination angle f 5 0°; long-dashedcircle, f 5 30°; dashed–dotted circles, f 5 75°, short-dashed cir-cles, Lambertian source.


~f!. A coefficient ~ma!, a scattering coefficient ~ms!,and an anisotropic factor ~g! of the homogeneoussample were chosen to be in the range of typicalhuman tissue.16 20 3 106 photons were launchedfor each simulation. We calculated the fluorescenceand reflectance at illumination angles of 0° to 75° atintervals of 15°. The normalized distributions of cal-culated fluorescence and reflectance at three repre-sentative illumination angles are displayed in Fig.1~b!. For clear illustration and discussion, only thedata in the incidence plane defined by the incidenceaxis and the normal of the tissue surface are dis-played. It is not a surprise that the fluorescence isalmost Lambertian, regardless of the illuminationangle, because the fluorescence source is isotropic.The reflectance has a distribution that is nearly Lam-bertian when the illumination is perpendicular to thetissue surface ~f 5 0°!. However, the symmetry ofthe reflectance is destroyed when the illuminationbecomes oblique. The major cause of the asymmetryis that tissue is the forward-scattering dominant tur-bid medium ~g ; 0.9, ms .. ma!. Backscatteringwith respect to the illumination axis is much lesslikely than forward scattering, which indicates thatthe diffusion and photon migration theories will pro-duce errors in the calculation of angularly resolvedreflectance because both theories treat fluorescenceand reflectance as diffusive sources. Taking an ap-proach similar to that of Profio2 and Lenz,4 we calcu-lated the ratio ~FyR! of fluorescence to reflectance ofthe excitation collected at the same angle ~u!. Theresults are shown in Fig. 2. The dashed–dottedsemicircle with a radius FyR angle u 5 f 5 0° isdisplayed as a reference. Comparing the FyR ratioin the various illumination-and-collection geometrieswith the reference, we can see clearly that, whenangle u is close to f, the ratio value varies little andthe FyR ratio effectively corrects the geometrical ef-fect. In contrast, the fluorescence emission shown inFig. 1~b! is highly dependent on collection angle u asa Lambertian source. Our results are consistentwith that stated by Wu et al.,13 though the reflectanceused in this FyR ratio method is at the wavelength ofthe excitation light instead of at that of the fluores-cence emission. It should be emphasized that thecorrection by the FyR ratio method becomes invalid

Fig. 2. Angular distribution of FyR ratio at several illuminationangles f in the incidence plane. Solid curve, f 5 0°; long-dashedurve, f 5 30°; short-dashed curve, f 5 75°; dashed–dotted curve,

reference.

ru

watBtnmicl

rcrcflc

dd

when F and R are collected at an angle far from theillumination axis, as we discuss in Section 5 below.

In a practical imaging system the fluorescence andreflectance signals are collected at a solid angle. Wecalculated the FyR ratio of a system that collects thesignals within a cone of a 65° angle from the illumi-nation axis. The collection geometry and results areshown in Fig. 3. As can be seen, the FyR ratioweakly depends on the illumination angle ~f!. Thecorrection of the geometrical effects is valid for a widerange of illumination angles.

It should be noted that the illumination-and-collection geometry in the model is not appropriatefor many applications, such as endoscopic imagingsystems. Also, the model was based on one set ofoptical properties of tissue. Nevertheless, the mod-eling results provide some insights into the relation-ship between fluorescence and excitation reflectance,and they seem encouraging. In principle, one canuse Monte Carlo modeling to study this FyR ratiomethod in any fluorescence excitation-and-collectiongeometry. However, the computation time and dataspace will be significantly increased when a geometrywith noncollimated illumination and an irregular tis-sue surface is modeled.15 The experimental investi-gation is more efficient for studying the relationbetween fluorescence and reflectance in a practicalgeometry such as an endoscopic imaging system.

As we pointed out at the beginning of this section,the reflection of excitation light from a tissue surfaceincludes three components. Although the diffuse re-flectance may be used as the reference in the FyRatio method to correct the geometrical effects, spec-lar reflection will create artifacts.2,4,13 Measures

must be taken to remove specular reflection from thereflection of the excitation light and retain the diffu-sive components. Tissue is a kind of turbid medium.Light is scattered and absorbed during propagationinside the tissue. In the visible region the light intissue is highly scattered and the transport of light isdominated by scattering events. It is well knownthat the polarization of light will be changed when ascattering event occurs. If the excitation light has acertain state of polarization, multiple scattering inthe tissue, which yields diffusive reflectance,will randomize the state of polarization. In contrast,

Fig. 3. FyR ratio collected by an imaging device such as an en-oscope. Left, illumination-and-collection geometry; right, depen-ence of FyR ratio on illumination-and-collection angle.

specular reflection maintains the original polariza-tion state of the excitation light. Thus apolarization-sensitive imaging system can separatethe specular and the diffusive reflectance compo-nents. Recently Demos and Alfano,17 Demos et al.,18

and Perelman et al.19 used the polarization imagingtechniques to characterize tissue structures.

To reject the specular reflection and extract thediffusive components from the total reflectance of ex-citation light we utilize a cross-polarization imagingmethod. Linearly polarized excitation was chosen forthis study because linearly polarized light is moreeasily randomized than is circularly polarized light ina turbid medium.20,21 Experimentally, we illumi-nate a tissue-simulating sample ~turbid medium!

ith linearly polarized fluorescence excitation lightnd by the use of a polarization analyzer in front ofhe imaging sensor record the cross-polarized image.ecause forward scattering dominates light propaga-

ion in the human tissue, the cross-polarized compo-ents in the reflection are produced mainly byultiple scattering and are highly diffusive. The

nformation carried in the cross-polarization imageomes predominantly from beneath the superficialayer.17–19 The specular reflection and superficial

information ~mainly as a result of single backscatter-ing! that cause artifacts in the correction of geomet-ical effects are almost completely removed in theross-polarized image. Taking the ratio of the fluo-escence signal to the cross-polarized reflection re-orded from the same position forms a normalizeduorescence image. The geometrical effects can beorrected in the FyR ratio image by the approach

reported in Ref. 13.

3. System and Materials

A. Experiment

The experimental arrangement for recording thecross-polarized and fluorescence images is shown inFigure 4. The excitation source was a solid-statefrequency-doubled laser with output power of ;200mW at a wavelength of 457 nm ~blue microlaser,Laser Power Corporation!. The laser light was de-

Fig. 4. Schematic diagram of the experimental setup used tocollect the fluorescence and cross-polarized reflection images: F1,long-pass filter; P1, P2, polarizers with polarization axes perpen-dicular to each other; u9, angle between the optical axis of theendoscope and the normal of the sample surface.


osc

riI~matsa;

tm

h

bofitattNtcfo3mesm

3

livered to the sample surface through an illuminationoptics consisting of a multimode optical fiber of200-mm diameter, a linear polarizer, and microlensesf 1.6 mm-diameter. The fluorescence and reflectionignals from the sample were collected by a standardommercial endoscope ~7200A, Karl Storz! and im-

aged to an 8-bit CCD camera ~WAT-502A, Watec,Inc.!. The imaging channel of the endoscope was;1.8 mm in diameter. The angle between the opti-cal axes of the endoscope and the illumination opticswas ;15°. The distal tips of the endoscope and theillumination optics were attached to each other andwere located ;10 mm away from the surface of thesample. The imaged area on the sample surface,which was smaller than the illuminated area, wasapproximately 10 mm 3 10 mm at u 5 0°. To varythe geometry of illumination and collection, the sam-ple could be rotated along the axis across point A andperpendicular to the plane of the page in Fig. 4. Theholder of a long-pass filter with a cutoff wavelength at470 nm and a polarizer with its polarization axisperpendicular to that of excitation light allows foreasily selecting the fluorescence or the cross-polarized imaging channel. The system extinctionratio of the cross-polarized imaging channel is;100:1. The extinction ratio is defined as the ratioof the maximal transmittance ~parallel! for a linearlypolarized light to the minimal transmittance ~cross!.The fluorescence and cross-polarized images aregrabbed by a frame-grabber ~Genesis-LC, MatroxElectronic Systems, Ltd.! at rate of 25 framesys. Toimprove the signal-to-noise ratio we formed the flu-orescence image by taking an average of 16 imageframes. The ratio image of fluorescence versuscross-polarized reflection was then generated by a PCcomputer.

The SNR of the ratio image is strongly dependenton the quality of the fluorescence and the cross-polarized images. We found that the speckle effectdid cause the quality of reflection image to deterio-rate because the excitation was from a laser, a coher-ent source. Many methods to reduce the speckleeffect and improve the image quality have beenproposed.22–24 Here we used a vibration mechanismeported in Ref. 22 to modulate the fiber and random-ze the phases of the propagating modes in the fiber.n our experimental setup a small part of the fiber;5 mm in length! was driven by a voice coil. Whenodulation was applied, the speckle pattern becametime-varying function that depended on the vibra-

ion amplitude and frequency. We found that thepeckle noise in the reflection image can be effectivelyveraged out when the modulation frequency is set to600 Hz.

B. Tissue Phantoms

Tissue-simulating phantoms were made of gelatinthat consisted of 20% solids dissolved in boilingdeionized water, polystyrene spheres of 0.55-mm di-ameter ~Polysciences, Inc.!, a fluorescent dye mix-ure, and blood, the dominant absorber. Theethod for preparing the gel-based solid phantom


as been well documented.25,26 The scattering coef-ficients of the tissue phantoms, which were controlledby the concentration of polystyrene microspheres,were determined by two independent methods:measurements of the total attenuation coefficientsand calculation by Mie theory.27–31 In the first ap-proach we measured the unscattered transmittanceof an optically thin scattering sample that was madefrom the gel–microsphere mixture without addingany absorbers. The total attenuation coefficient, thescattering coefficient of a pure scattering sample,then could be obtained from the unscattered trans-mittance by the Beer–Lambert law. We used thespectrophotometer shown in Fig. 5 to determine theoptical properties of the tissue phantoms. We usedan optical fiber of 200-mm diameter to conduct broad-and illumination from a white-light source ~fiberptic illuminator 77501, Oriel!. The light from theber was collimated into a beam of 2-mm diameter byhe use of a lens and three apertures. The divergentngle of the beam was approximately 10 mrad. Theransmission signal was collected by a lens–fiber sys-em and analyzed by a spectrometer ~MCS 55 UV-IR, Zeiss!. The thickness of the samples used in

he measurements varied from 0.3 to 0.45 mm. In ouralculation of scattering coefficient and anisotropicactor ~g! by Mie theory we used the refractive indicesf gel and polystyrene microspheres reported in Refs.1 and 32. Figure 6~a! displays the experimentallyeasured and theoretically calculated scattering co-

fficients of the tissue phantoms used in this re-earch. The experimental scattering coefficient of aicrosphere concentration at 0.5% by weight ~wyw!

was extrapolated from the data at 0.25% and 0.35%wyw because the spectrometer was not sensitiveenough to detect highly attenuated light. As can beseen, the experimental results are well consistentwith the calculation of Mie theory. The calculated gfactor is shown in Fig. 6~b!. The scattering coeffi-cients and anisotropic factor of the tissue phantomswere in the range of typical human tissue.16

The amount of absorption of tissue phantoms isdetermined by the concentration of blood, the domi-nant absorber. In this study, fresh blood was drawnfrom mice. Red cells were rinsed with phosphate-buffered saline ~pH 7.4! and then lysed by addition ofdeionized water. Fragments of cell membraneswere removed by centrifugation. The absorption co-

Fig. 5. Setup for measurement of the optical properties of thesamples.

mdsttit

isgtiatptmcmnAia

tojial

ptt

t

efficient of the hemolyzed blood was measured withthe spectrophotometer shown in Fig. 5. The concen-tration of hemoglobin ~most cells were oxygenated!could be determined by the ratio of measured absorp-tion coefficient to that of pure oxyhemoglobin re-ported in Refs. 33 and 34. By controlling preciselythe hemoglobin concentration in the gel, we set thecontent of blood with a hematocrit of 0.51 for thethree groups of tissue phantoms that we used to 2.5%,5%, and 7.5% by volume ~vyv!.

To simulate fluorophores in the tissue we used aixture of highly efficient green and red Stabilo Boss

ye fluorescent highlighters. A typical fluorescencepectrum of a tissue phantom measured by a spec-rometer is shown in Fig. 7. The fluorescence spec-ral coverage of the dye mixture is also in the range ofn vivo measured autofluorescence of humanissues.1–12 The absorption of the fluorescence dyes

at the concentration used in the tissue phantoms ismuch less than that of blood and can be negligible.

Fig. 6. Scattering coefficients and g factors of tissue-simulatinghantoms used in the study. ~a! Measured and calculated scat-ering coefficients at concentrations of polystyrene microsphere inhe tissue phantom of 0.25%, 0.35%, and 0.5% wyw. ~b! Calcu-

lated g factors of 0.55-mm-diameter polystyrene microspheres inhe gel.

All samples used in this study were ;25 mm in di-ameter and 4.5 mm thick.

4. Results

A. Homogeneous Phantoms

We demonstrate that the geometrical effects on fluo-rescence imaging of the homogeneous tissue phantomcan be corrected to a great extent by the FyR ratiomaging method. In the first experiment, all theamples with different optical properties were homo-eneous and had flat surfaces. The compositions ofhe samples are listed in Table 1. We changed themaging geometry by rotating the sample along thexis across point A and perpendicular to the plane ofhe page in Fig. 8. The fluorescence and cross-olarized images were collected from all tissue phan-oms at u9 5 0°–60° in increments of 15°. For theeasurements at large angles u, the illumination be-

ame highly nonuniform. We did not perform aeasurement for u9 . 60° because of the limited dy-amic range and sensitivity of our 8-bit CCD camera.lso, the image blurring caused by aberrations in the

maging system became the major source of error asngle u became greater than 60°.For a fair comparison, the mean gray levels of all

he raw fluorescence images and ratio images of flu-rescence against cross-polarized reflection were ad-usted to 128, half of the full gray levels of an 8-bitmage. The typical raw fluorescence and ratio im-ges are shown in Fig. 9. The histogram for the grayevel in the figure indicate the homogeneity of the

Fig. 7. Fluorescence spectrum of a tissue phantom excited by alaser of 457-nm wavelength.

Table 1. Composition of Homogeneous Phantoms with Flat Surfaces

SampleNumber

Blood Content~vyv, %!

Concentration of PolystyreneMicrosphere ~wyw, %!

1 2.5 0.252 5.0 0.253 7.5 0.254 2.5 0.55 5.0 0.56 7.5 0.5


bmw


image. As can be seen from Fig. 9, the raw fluores-cence image is highly nonuniform and the gray levelsvary across a wide range. In contrast, the ratio im-age is uniform and its gray level has a narrow distri-bution about 128, the mean gray level. We foundthat the standard deviations of gray levels for the rawfluorescence images recorded from all six phantomsvaried from 24 to 41 at u9 5 0°–60°, whereas thestandard deviations of the ratio images fell into arange from 1.8 to 4.5, much smaller than that for theraw fluorescence images.

In the second experiment we examined the perfor-mance of the ratio technique for imaging of a homo-geneous sample with an irregular surface. Theprofile of the phantom surface was constructed asshown in Fig. 8. The width and the depth of thegrooves were ;8 mm and ;5 mm, respectively. Thelood content and the concentration of polystyreneicrosphere of the phantom used in the experimentere 5% vyv and 0.35% wyw, respectively. The flu-

orescence and cross-polarization images were takenin the same fashion as for the first experiment at u9 5

od content of 2.5% ~vyv! and a microsphere concentration of 0.5%ray level, ~b! FyR ratio image and its histogram.

Fig. 8. Simplified geometry of the experimental imaging system.Left, Imaging of a flat surface phantom at u9 Þ 0; right, imaging ofa phantom with an irregular surface at u9 5 0°. f is theexcitation-and-collection angle.

Fig. 9. Images collected from a tissue-simulating phantom with a blo~wyw! at u9 5 30°. ~a! Raw fluorescence image and histogram of its g

rv s

0°. The raw fluorescence and ratio images areshown in Fig. 10. As can be seen, the inhomogeneityin the raw fluorescence has been significantly com-pensated for by the FyR ratio imaging technique.

B. Inhomogeneous Phantoms

In this study we evaluated the performance of theFyR ratio imaging on inhomogeneous tissue phan-toms. We classified the phantoms based on theamount of inhomogeneity in the concentration of flu-orophores and optical properties. To construct aninhomogeneous tissue phantom we replaced parts ofthe homogeneous sample with samples of differentfluorophore concentrations or optical properties.For the fluorescence contrast phantom shown in Fig.11~a! the parts of the homogeneous sample in theshaded areas were replaced with simulated lesions ofslightly lower fluorescent dye concentration and thesame optical properties. The blood content and themicrosphere concentration of the phantom were set to5% and 0.35%, respectively. Two shallow holes weremade to make the surface irregular and to creategeometrical interference for identifying the simu-lated lesions. The structure of the absorption andscattering contrast phantoms is shown in Fig. 11~b!.For the absorption contrast phantom the parts of thesample in areas A and B were replaced with samplesof 10% and 2.5% blood content, respectively. Theblood content of the surrounding sample was 5%.The concentration of polystyrene microspheres in allsamples that we used to construct the absorptioncontrast phantom was set to 0.35%. For the scatter-ing contrast phantom, samples in areas A and B were

Fig. 10. Raw fluorescence image and ratio image recorded fromthe homogeneous phantom of an irregular surface. ~a! Raw fluo-escence and the profile across A–A9, ~b! ratio image of fluorescenceersus cross-polarized reflection and profile across A–A9. The

standard deviations of the gray levels of the raw fluorescence andratio images are 32 and 5, respectively.

replaced with samples of microsphere concentrations0.5% and 0.35%, respectively. The microsphere con-centration of the surrounding sample was 0.25%.The blood content in all samples for the constructionof the scattering contrast phantom was set to 5%.The concentrations of fluorescent dyes in all the ab-sorption and scattering contrast phantoms were con-stant.

The results shown in Fig. 12 are the raw fluores-cence and ratio images of three inhomogeneous phan-toms. The back-to-back comparison of the rawfluorescence and ratio images demonstrates againthat the geometrical effects have been well correctedin the FyR ratio method. The image of the fluores-cence yield contrast phantom demonstrates that theratio technique can clearly identify the simulated le-sions that yield slightly lower fluorescence. In theraw fluorescence image, one cannot separate the ar-tifacts created by two holes from the simulated le-sions. However, only the simulated lesions thatyield low fluorescence appear clearly in the ratio im-age, and the geometrical artifacts caused by the holeswere almost completely eliminated. Furthermore,the ratio image of the absorption contrast phantomshows that the variation of fluorescence signal causedby the uneven distribution of blood has been compen-sated for effectively. The results from the measure-ment of the scattering contrast phantom suggest thatthe variation of scattering property across the tissuesurface can be detected in the FyR ratio image.

5. Discussion

We found that specular reflection in the cross-polarized image was significantly reduced by a cross-polarization imaging system with an extinction ratio

Fig. 11. Structures of inhomogeneous tissue-simulating phan-toms: ~a! fluorescence yield contrast phantom, ~b! absorption andcattering contrast phantoms.


3

of 100:1. The image shown in Fig. 13~a! is a reflec-tion image taken from the fluorescence yield contrastphantom at u9 5 0° with the polarizer in the cross-polarization imaging channel removed. The specu-lar reflection signal in the image is much higher thanthat of diffusive reflection and a saturated large areaof CCD sensor. Comparing the FyR images in Fig.13 with that in Fig. 12~b! makes clear that specularreflection causes major artifacts if the direct reflec-tion signal is used as the reference to compensate forthe geometrical effects in the FyR ratio technique.We could further improve the performance of the sys-tem used in this study by optimizing the optical de-sign and increasing the extinction ratio.

In the measurements of the homogeneous phantomwith a flat surface, the homogeneity of the ratio im-age decreased slightly as angle u9 increased but wasnot sensitive to the polarization plane of the excita-tion light and to the optical properties of samples.In the measurements of the homogeneous phantomwith an irregular surface, the line profile analyses inFig. 10 show that a small variation of the pixel valuein the ratio image does have a weak correlation with

Fig. 12. Results of imaging of inhomogeneous phantoms. ~a! andcontrast phantom. Figure 12 continues on the next page.


the profile of the sample surface. These results in-dicate that the excitation incident angle and the flu-orescence collection angle play roles, thoughinsignificant ones, in determining the FyR ratio valueobtained. The simplified geometry of our imagingsystem is shown in Fig. 8. The excitation incidentangle and the fluorescence collection angle are as-sumed to be the same because the sizes of the illu-mination microlens and the endoscope lens are muchsmaller than the distance from the sample to thedistal tips of the endoscope and the illumination op-tics. The range of excitation-and-collection angle ffor a phantom with a flat or irregular surface is thendetermined by the degree of irregularity in the sur-face and by angle u9. In Fig. 14 we display the pixelvalue of ratio images as a function of angle f based onmeasurements of all homogeneous tissue phantoms.The value of the ratio increases slightly as f in-creases in the range 0–70°. This result seems con-sistent with the results of Monte Carlo modelingshown in Fig. 3, although the experimental and mod-eling geometries for endoscopic imaging are different.The ratios at angle f . 70° were not measured for the

aw fluorescence and ratio images taken from a fluorescence yield
~b! R

adcssbmcofl

csatt0saTnuafsttcnam

reasons discussed in Section 3. We found that themaximal variation of ratio in the range of angle ffrom 0 to 70° is less than 7%. This sets the accuracyor sensitivity in relative fluorescence mapping of tis-suelike turbid media.

The results of Monte Carlo modeling in Fig. 2 showthat the FyR ratio remains constant and that thegeometrical correction is valid for F and R collected ina specific range of backscattering angle ~calculated asuf 2 uu! from the illumination axis. Angles u and fre defined in Fig. 1~a!. However, the ratio valuerops quickly when backscattering angle uf 2 uu be-omes large. This shows that the reflectance has aimilar angular distribution to the fluorescence for amall backscattering angle from the illumination axisut not for a large angle. The difference is causedainly by the contribution of the nearly diffusive

omponent that is the result of photons scatterednce or twice in the tissue35 to the total diffuse re-ectance, which consists of both a nearly diffusive

Fig. 12. ~c! and ~d!. Raw fluorescence and ratio images of the a

omponent and a diffusive component. The diffu-ive component is produced by multiple scatteringnd has characteristics similar to those of Lamber-ian fluorescence. As light propagation in tissuelikeurbid media is dominated by forward scattering ~g ;.9!, the probability of a single-scattering event in amall backscattering angle is much smaller than forlarge backscattering angle and forward scattering.hus the nearly diffusive reflectance must be domi-ated by photons backscattered in a large angle ~uf 2u! from the illumination axis. In contrast, small-ngle backscattering from tissue is dominant in dif-usive reflectance, which has an angular distributionimilar to that of Lambertian fluorescence. The con-ribution of the nearly diffusive component to theotal diffuse reflectance becomes significant with in-reasing backscattering angle uf 2 uu. Therefore theearly diffusive photons in the total reflectance causedecrease in the FyR ratio and invalidate the geo-etrical correction in a large backscattering angle

tion contrast phantom. Figure 12 continues on the next page.
bsorp

3

uf 2 uu. For good correction of the geometrical effectthe fluorescence and the diffuse reflectance must becollected in a small backscattering angle in which thecontribution of the diffusive component to the totaldiffuse reflectance is dominant. Apparently, the en-doscopic imaging system used in the experimentalstudy satisfies that condition.

The measurement of the fluorescence contrastphantom demonstrates that the ratio imagingmethod can identify simulated lesions that have afluorescence that is slightly lower than that of thesurrounding sample and can eliminate geometricalartifacts. Quantitatively, the fluorescence of thesimulated lesions used in the experiment was set to;80% of that of the surrounding samples. In rawfluorescence it was not possible to distinguish thelesions from the artifacts created by two shallowholes on the surface of the sample. In the ratio im-age the difference in fluorescence between simulated

Fig. 12. ~e! and ~f ! Raw fluorescence and r


lesions and simulated normal tissue was obvious.Line profile analysis has shown that the border be-tween the simulated lesions and the simulated nor-mal tissue is smooth because of the migration offluorescence photons from the surrounding samples

Fig. 13. Left, reflection image recorded without the cross polar-izer; right, FyR ratio image without the cross polarizer.

images of the scattering contrast phantom.
atio

irecfwcduTacydvtp

to the simulated lesion areas. A clinical studyshowed that the contrasts in fluorescence betweenearly lesions and normal tissue in many organ sitesare greater than in our phantom.5–8,10 Our resultshold the promise for potential application to identi-fying early lesions clinically by the use of the ratioimaging technique.

It is an important finding that the ratio method caneffectively correct the artifacts caused by variation ofblood content in the sample. The blood in tissue isusually not evenly distributed, depending on local vas-cularization. As a dominant absorber, blood stronglyaffects the distribution of excitation light in tissue.The fluorescence intensity of tissue is correlated tolocal blood content, as shown in the raw fluorescenceimage recorded from the absorption contrast phantom.A variation of blood content also appears in the cross-polarized reflection image. However, the FyR ratiomage indicates that the effects of absorption on fluo-escence and diffuse reflectance signals can cancelach other. In the measurement of the scatteringontrast phantom we found that the reflectance is af-ected by variation of the scattering coefficient,hereas the fluorescence signal is less sensitive to

hanges in scattering. It appears that the unevenistribution of scatterers does not significantly contrib-te to the inhomogeneity of a raw fluorescence image.he geometrical effects are corrected in the ratio im-ge, but the high and low scattering areas appearlearly and create artifacts in maps of fluorescenceield. Fortunately, the structure of real tissue, whichetermines the scattering coefficients, is not likely toary so suddenly as in the scattering contrast phan-om. The artifacts of scattering may not play an im-ortant role in clinical applications.

6. Conclusions

A combination of cross-polarized and fluorescence im-aging techniques has been successfully used to cor-rect geometrical effects in fluorescence measurementof tissuelike turbid media. This technique permits

Fig. 14. Relative change of FyR ratio as a function of excitationand collection angle u. The vertical axis is the difference betweenthe ratio at angle u and at 0° divided by the ratio at 0°.

mapping of fluorescence yield over a homogeneoustissue surface. The sensitivity and accuracy of thedetection of variations in fluorescence depend weaklyon the excitation incident and fluorescence collectionangles. Based on the experimental results, the ratioimaging technique can identify a variation of fluores-cence in tissue of greater than 7% when excitationincident and fluorescence collection angles are in therange 0–70°. It should be emphasized that the the-oretical analysis and the experimental study weredone with single-layered tissuelike samples. As wasdiscussed in Ref. 13, real tissues are usually multiplylayered; their fluorophores are distributed as a func-tion of depth. Extensive in vivo investigation will beneeded to evaluate the performance of this ratio tech-nique for clinical practice. Our future research willfocus on understanding the relation between polar-ized light and fluorescence in tissue and on construct-ing an endoscopic imaging system based on the ratiomethod for in vivo study of autofluorescence of hu-man tissue.

This research was funded by Hong Kong Re-search Grants Council grants CA97y98.EG01 andHKUST6207y98P and by the Hong Kong Universityof Science and Technology.

References1. R. R. Alfano, D. B. Tat, J. Cordero, P. Tomashefsky, F. W.

Longo, and M. A. Alfano, “Laser induced fluorescence spectros-copy from native cancerous and normal tissue,” IEEE J. Quan-tum Electron. QE-20, 1507–1511 ~1984!.

2. A. E. Profio, D. R. Doiron, and J. Sarnaik, “Fluorometer forendoscopic diagnosis of tumors,” Med. Phys. 11, 516–520~1984!.

3. C. Kittrell, P. L. Willett, C. de los Santos-Pacheo, N. B. Ratliff,J. R. Kramer, E. G. Malk, and M. S. Feld, “Diagnosis of fibrousarterial atherosclerosis using fluorescence,” Appl. Opt. 24,2280–2281 ~1985!.

4. P. Lenz, “Endoscopic fluorescence detector,” Rev. Sci. Instrum.59, 930–933 ~1988!.

5. R. M. Cothren, R. R. Richards-Kortum, M. V. Sivak, M. Fitz-maurice, R. P. Rava, G. A. Boyce, G. B. Hayes, M. Doxtader, R.Blackman, T. Ivanc, M. S. Feld, and R. E. Petras, “Gastroin-testinal tissue diagnosis by laser-induced fluorescence spec-troscopy at endoscopy,” Gastrointest. Endosc. 36, 105–111~1990!.

6. J. Hung, S. Lam, J. C. LeRiche, and B. Palcic, “Autofluores-cence of normal and malignant bronchial tissue,” Lasers Surg.Med. 11, 99–105 ~1991!.

7. N. Ramanujam, M. F. Mitchell, A. Mahadevan, S. Thomsen, E.Silva, and R. Richards-Kortum, “Fluorescence spectroscopy:a diagnostic tool for cervical intraepithelial neoplasia ~CIN!,”Gynecol. Oncol. 52, 31–38 ~1994!.

8. J. Qu, C. MacAulay, S. Lam, and B. Palcic, “Laser-inducedfluorescence spectroscopy at endoscopy: tissue optics, MonteCarlo modeling, and in vivo measurements,” Opt. Eng. 34,3334–3343 ~1995!.

9. S. Warren, K. Pope, Y. Yazdi, A. J. Welch, S. Thomsen, A. L.Johnston, M. J. Davis, and R. Richards-Kortum, “Combinedultrasound and fluorescence spectroscopy for physico-chemicalimaging of atherosclerosis,” IEEE Trans. Biomed. Eng. 42,121–132 ~1995!.

10. T. D. Wang, J. van Dam, J. M. Crawford, E. A. Preisinger, Y.Wang, and M. S. Feld, “Fluorescence endoscopic imaging of


human colonic adenomas,” Gastroenterology 111, 1182–1191

1

1

1

1

1

1

1

1

1

2

2

2

2

G. R. Jones, “Chromatic modulation for optical fiber sensing

2

2

2

2

3

~1996!.1. G. A. Wagnieres, A. P. Studzinski, and H. E. van den Bergh,

“An endoscopic fluorescence imaging system for simultaneousvisual examination and photodetection of cancers,” Rev. Sci.Instrum. 68, 203–212 ~1997!.

2. K. Tumer, N. Ramanujam, J. Ghosh, and R. Richards-Kortum,“Ensembles of radial basis function networks for spectroscopicdetection of cervical precancer,” IEEE Trans. Biomed. Eng. 45,953–961 ~1998!.

3. J. Wu, M. S. Feld, and R. P. Rava, “Analytical model for ex-tracting intrinsic fluorescence in turbid media,” Appl. Opt. 32,3585–3595 ~1993!.

4. C. M. Gardner, S. L. Jacques, and A. J. Welch, “Fluorescencespectroscopy of tissue: recovery of intrinsic fluorescence frommeasured fluorescence,” Appl. Opt. 35, 1780–1792 ~1996!.

5. L. Wang and S. L. Jacques, “Monte Carlo modeling of lighttransport in multilayered tissue in standard C” ~University ofTexas, Houston, Tex., 1992!.

6. W. F. Cheong, S. A. Prahl, and A. J. Welch, “A review of theoptical properties of biological tissues,” IEEE J. QuantumElectron. 26, 2166–2185 ~1990!.

7. S. G. Demos and R. R. Alfano, “Optical polarization imaging,”Appl. Opt. 36, 150–155 ~1997!.

8. S. G. Demos, A. J. Papadopoulos, H. Savage, A. S. Heerdt, S.Schantz, and R. R. Alfano, “Polarization filter for biomedicaltissue optical imaging,” Photochem. Photobiol. 66, 821–825~1997!.

9. L. T. Perelman, V. Backman, J. Wu, R. R. Dasari, and M. S.Feld, “Spectroscopic diagnostics of epithelial tissues with po-larized light,” in Optical Tomography and Spectroscopy of Tis-sue III, B. Chance, R. R. Alfano, and B. J. Tromberg, eds., Proc.SPIE 3597, 474–479 ~1999!.

0. J. M. Schmitt, A. H. Gandjbakhche, and R. F. Bonner, “Use ofpolarized light to discriminate short-path photons in a multi-ply scattering medium,” Appl. Opt. 31, 6535–6546 ~1992!.

1. G. D. Lewis, D. L. Jordan, and P. J. Roberts, “Backscatteringtarget detection in a turbid medium by polarization discrimi-nation,” Appl. Opt. 38, 3937–3944 ~1999!.

2. S. Jutamulia and T. Asakura, “Reduction of coherent noise usingvarious artificial incoherent sources,” Optik 70, 52–57 ~1985!.

3. R. Smith, S. Ahmad, A. Voudas, J. Spencer, A. Russel, and


electromagnetic and speckle noise analysis,” J. Mod. Opt. 39,2301–2314 ~1992!.

4. T. Iwai and T. Asakura, “Speckle reduction in coherent infor-mation processing,” Proc. IEEE 84, 765–781 ~1996!.

5. B. W. Pogue, L. Lilge, M. S. Patterson, B. C. Wilson, and T.Hasan, “Absorbed photodynamic dose from pulsed versus con-tinuous wave light examined with tissue-simulating dosime-ters,” Appl. Opt. 36, 7257–7269 ~1997!.

6. L. Wang and X. Zhao, “Ultrasound-modulated optical tomog-raphy of absorbing objects buried in dense tissue-simulatingturbid media,” Appl. Opt. 36, 7277–7282 ~1997!.

7. S. L. Jacques, C. A. Alter, and S. A. Prahl, “Angular depen-dence of HeNe laser light scattering by human dermis,” LasersLife Sci. 1, 309–333 ~1987!.

28. W. Pickering, S. A. Prahl, N. van Wierington, J. F. Beek,H. J. C. M. Sterenborg, and M. J. C. van Gemert, “Double-integrating-sphere system for measuring the optical propertiesof tissue,” Appl. Opt. 32, 399–410 ~1993!.

29. J. Qu, C. MacAulay, S. Lam, and B. Palcic, “Optical propertiesof normal and carcinomatous bronchial tissue,” Appl. Opt. 33,7397–7405 ~1994!.

30. A. J. Drukin, S. Jaikumar, and R. Richards-Kortum, “Opticallydilute, absorbing, and turbid phantoms for fluorescence spec-troscopy of homogeneous and inhomogeneous samples,” Appl.Spectrosc. 47, 2114–2121 ~1993!.

31. M. Kohl, M. Essenpreis, and M. Cope, “The influence of glucoseupon the transport of light in tissue-simulating phantoms,”Phys. Med. Biol. 40, 1267–1287 ~1995!.

32. D. W. Leonard and K. M. Meek, “Refractive indices of thecollagen fibrils and extrafibrillar material of the corneal stro-ma,” Biophys. J. 72, 1382–1387 ~1997!.

33. E. J. van Kampen and W. G. Zilstra, “Determination of hemo-globin and its derivatives,” in Advances in Clinical Chemistry,H. Sobotka and C. P. Stewart, eds. ~Academic, New York,1965!, Vol. 8, pp 158–187.

34. O. W. van Assendelft, Spectrophotometry of haemoglobin de-rivatives ~Royal Vangorcum, Assen, The Netherlands, 1970!.

35. L. Wang and S. L. Jacques, “Analysis of diffusion theory andsimilarity relations,” in Photon Migration and Imaging inRandom Media and Tissues, B. Chance and R. R. Alfano, eds.,Proc. SPIE 1888, 107–116 ~1993!.

Documents

Excitation-and-collection geometry insensitive fluorescence imaging