Upload
nicholas-paul
View
217
Download
0
Embed Size (px)
Citation preview
Digital Pathology Solutions Conference [email protected]
TOWARD A DIAGNOSIS TOWARD A DIAGNOSIS ASSISTANCE SYSTEM FOR ASSISTANCE SYSTEM FOR
DIGITAL PATHOLOGY OF BREAST DIGITAL PATHOLOGY OF BREAST CANCERCANCER
M. Oger, P. Belhomme, J.J. Michels, A. Elmoataz
GRECAN, EA 1772,University of Caen Basse-NormandieF. BACLESSE Cancer Centre, CaenGREYC, UMR 6072, University of Caen Basse-Normandie
Digital Pathology Solutions Conference [email protected]
IntroductionIntroduction
• Identification of breast tumor lesions is not always a easy task.
• Cancer lesions are sometimes heterogeneous.
• Question: is automatic image processing able to help classifying benign and malignant breast lesions?
Digital Pathology Solutions Conference [email protected]
AimAim
• To try to develop automatedComputer-Aided Diagnosis (CAD) toolsfor pathologists
• To work with Virtual Slides (VS) in order to take into account lesion heterogeneity
Digital Pathology Solutions Conference [email protected]
Material and methodMaterial and method
• Low resolution Virtual Slide6 µm: Nikon CoolScan 8000 ED.
• 224 images (different size) are included in the knowledge base
• 28 histological types• 3 histological families (Benign, Malignant Carcinoma,
Malignant Sarcoma)
slide holder
images with foci of different histological type exist, but we labeled them according to the dominant type
Digital Pathology Solutions Conference [email protected]
Example of low resolution VSExample of low resolution VS
• At the resolution of 6 µm, pathologists recognize fairly easily histological types in 80 to 90% of cases.
but “small objects” are sometimes difficult to identify
Fibroadenoma Intraductal carcinoma
2228 X 1915 px = 12.3 Mb3479 X 2781 px = 28 Mb
Digital Pathology Solutions Conference [email protected]
Material and methodMaterial and method
• A “new image” will be compared to the knowledge database.
• A graphical user interface will be built to allow a “visual” presentation of the results obtained.
Digital Pathology Solutions Conference [email protected]
• Multiparametric Analysis CAD system 1st version
• Spectral Analysis CAD system 2nd version
• Multiparametric Analysis CAD system 1st version
• Spectral Analysis CAD system 2nd version
Strategy ExplorationStrategy Exploration
Digital Pathology Solutions Conference [email protected]
Multiparametric analysisMultiparametric analysis
• We have developed a system which statistically determines the “similarity degree” of a new image compared to the different histological types.
• Requirements: » No segmentation
» Exploration of several color spaces: RGB, YCH1CH2 (Carron), AC1C2 (Faugeras), I1I2I3 (Ohta)...
• Application:» Computing a “signature” of parameters of the whole VS
» Comparing the signatures
Digital Pathology Solutions Conference [email protected]
The color signaturesThe color signatures• 234 global parameters computed on 6 color spaces
– Histograms– Mean– Median– Kurtosis– Skewness…
• + 13 "texture" parameters– S/N measure– Haralick…
• Vector distance (comparison of signatures) – Kullback-Leibler distance
• Software development– PYTHON language
n
i x
yy
y
xxdivKL
1
log.log.
Principal Component Analysis 188
Digital Pathology Solutions Conference [email protected]
► Automated systemAutomated system► InputInput = a new image = a new image► OutputsOutputs = similar = similar
imagesimagesfrom the knowledge from the knowledge basebase
CAD 1st version CAD 1st version systemsystem
Digital Pathology Solutions Conference [email protected]
Rank of the first image of the
same type
11 13.99 %13.99 %
≤ ≤ 33 33.33 %33.33 %
≤ ≤ 55 47.74 %47.74 %
≤ ≤ 1010 67.08 %67.08 %
Exhaustive analysis of the image database (one image vs the 223 others)with Kullback-Leibler distance
CAD 1st version: CAD 1st version: ResultsResults
Digital Pathology Solutions Conference [email protected]
CommentsComments
• Low resolution image classification is possible butthis strategy is a crude one which can lead only to a “preclassification” of the lesion under study
• Other strategies are to be explored
Digital Pathology Solutions Conference [email protected]
Strategy ExplorationStrategy Exploration
• Multiparametric Analysis CAD system 1st version
• Spectral Analysis CAD system 2nd version
Digital Pathology Solutions Conference [email protected]
Principle of spectral techniques Principle of spectral techniques for structural analysis of an for structural analysis of an
image databaseimage database
• Working on images with identical size• Comparing “point to point” each image with all
those of the database ==> the signature is the WHOLE image
• Trying to determine a “distance” between all the images of the database by using techniques of Spectral Dimensionality Reduction
• Replacing a n-dimensional space by a2D-visualization space (φ1, φ2)
Digital Pathology Solutions Conference [email protected]
Application to breast lesionsApplication to breast lesions• Problem:
– Database images are of various size
– In an image, some areas are uninformative (stroma, normal tissue, adipose cells...)
• Proposed solution: – Finding the interesting
“PATCHES” which describe the histological type at best
– Choosing an adequate size for “patches”: 32x32 px²
Digital Pathology Solutions Conference [email protected]
Example of 4 distinct classesExample of 4 distinct classes
• We work with:– Intra Ductal Carcinoma– Invasive Lobular Carcinoma– Colloid Carcinoma– Fibroadenoma
• We take only the 3 most representative VS of each class(□) 12 VS among 73
Invasive Lobular Carcinoma
Intra Ductal Carcinoma
Fibroadenoma
Colloid Carcinoma
Digital Pathology Solutions Conference [email protected]
IDC FA
ILC
CC
250 x 3 x 4 = 3000 retained patches
250 patches from each VS250 patches from each VS
Digital Pathology Solutions Conference [email protected]
Graph of the Graph of the selectedselected 4 types 4 types
Invasive Lobular Carcinoma
Fibroadenoma
Colloid Carcinoma
Intra Ductal Carcinoma
1 cross per patch = 3000 crosses
Digital Pathology Solutions Conference [email protected]
How can we analyseHow can we analysea a ““new imagenew image””
• 1) elimination of the background
Digital Pathology Solutions Conference [email protected]
• 2) Cutting in 32x32 patches
Digital Pathology Solutions Conference [email protected]
• 3) « patches » are projected on a 2D space (φ1, φ2)
φ1 = 0
Digital Pathology Solutions Conference [email protected]
• 4) segmentation by spectral analysis:patches corresponding to stroma are removed (cellular zones are preserved)
Stroma Cellular zones
φ1 = 0
Digital Pathology Solutions Conference [email protected]
Visual control
• 4) segmentation by spectral analysis:patches corresponding to stroma (Green) are removed, cellular zones (Purple) are preserved
Digital Pathology Solutions Conference [email protected]
CAD 2nd versionCAD 2nd version • 5) cellular patches
of the new image are projected onto the graph of cellular patches of the 4 histological types
Insertion of the new image
Digital Pathology Solutions Conference [email protected]
CAD 2nd versionCAD 2nd version
Intra Ductal Carcinoma 42,37%
Invasive Lobular Carcinoma 5,64%
Colloid Carcinoma 29,98%
Fibroadenoma 22,01%
Matching probabilities
2-neighborhood k-neighborhood
Results of a test done with a “new image” corresponding to an
Intraductal Carcinoma
Detail of the whole graph
Digital Pathology Solutions Conference [email protected]
ConclusionConclusion
• Technique of spectral analysis seems to be promising regarding 4 classes of tumors.
• This technique could be applied in order to try to identify tumor foci of different types on a virtual slide.
Digital Pathology Solutions Conference [email protected]
PerspectivesPerspectives
• But a lot of work remains to be done:– Extending the spectral analysis to 28 classes (the rest
of the database): improving the separation of the influence zone of each histological type.
– Increasing the signature: image patch + parameters which have been selected in the first part.
– Testing a higher resolution (sub sampled high resolution virtual slides).
Remark: the final strategy will be easily applicable to other tumor locations
Digital Pathology Solutions Conference [email protected]
Acknowledgements:
The authors gratefully acknowledge
Dr Paulette Herlin, Dr Benoît Plancoulaine,Dr Jacques Chasle,
the Regional Council of "Basse-Normandie"
and the "Comité départemental du Calvados de la Ligue de Lutte Contre le Cancer".