prostate cancer cad michael feldman, md, phd [email protected] assistant professor...
TRANSCRIPT
![Page 1: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/1.jpg)
Prostate Cancer CAD
Michael Feldman, MD, [email protected]
Assistant Professor PathologyUniversity Pennsylvania
![Page 2: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/2.jpg)
Outline
Automated image processing of whole slides
Spectral imaging for advanced tissue analysis of IHC/IF
![Page 3: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/3.jpg)
Overlap
CADSegmentationClassification
TissueCellular
Subcellular
MSIProtein Expression
Multiple probesTissueCellular
Subcellular
Diagnosis, Prognosis, Validation
![Page 4: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/4.jpg)
CAD – Computer Assisted Dx
Can we develop software to identify Prostate Cancer in whole digitized slides?
Utility Disease level Segmentation for IHC/IF, CAD for Rescreening or Primary Screening
![Page 5: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/5.jpg)
CAD is possible today Scanner’s
Image near optical diffusion limits 3rd and 4th generation scanners fast
Slides are large – 3-6 Gigabytes/slide Too large to use for algorithm
development Use many slides Analyze many features (Bayesian
approach) Feature combination
![Page 6: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/6.jpg)
Introduction
Most of the image is benign Can we reduce image size by excluding benign
areas Combine this with scale information to greatly
increase the efficiency of the detection procedure
![Page 7: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/7.jpg)
Find areas of interest at low scale (0.1 – 1X)
Only focus on these areas at high scale (2-40X)
Mimics human process
Scale (Magnification)
![Page 8: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/8.jpg)
Introduction
Each “pass” of the image rejects pixels, and then only the positive pixels are analyzed at higher scales
This allows us to efficiently analyze the image at higher scales by only looking at “interesting” pixels
Increasing accuracy does NOT increase execution time
![Page 9: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/9.jpg)
Introduction
Viola and Jones [1] introduced the “attention cascade” By quickly rejecting image regions that are definitely
negative using a minimum of features, we can concentrate computation on regions that might be positive
![Page 10: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/10.jpg)
Methods
![Page 11: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/11.jpg)
Methods To describe each pixel, we extract
~ 600 image features from the image There are three categories of features:
Statistical Features Co-occurrence (texture) Features Gabor Filter Features
Each of these is extracted from the three channels of the image, at three different window sizes
![Page 12: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/12.jpg)
Methods
Average Gabor Filter
Sum Variance Measurement of Correlation
![Page 13: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/13.jpg)
Methods Probability density
functions are constructed using the ground truth pixels provided by a pathologist
These PDFs are used to assign a likelihood value to each pixel based on the feature value Standard Deviation
![Page 14: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/14.jpg)
Methods The individual feature likelihoods are
combined by AdaBoost [3] to get a likelihood ensemble
A small number of features are used at scale 0 to obtain the ensemble, which is then thresholded
The process begins over again at the next scale; only pixels labeled as positive at the previous scale are analyzed
![Page 15: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/15.jpg)
Results
Graph of the computation time used to visualize efficiency in using the cascade
Visual comparison of the likelihood ensembles with the tumor mask
Receiver operating characteristic (ROC) curves were used to measure accuracy
Scatter plots visualize the separation of classes
![Page 16: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/16.jpg)
Results
![Page 17: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/17.jpg)
ResultsOriginal Image Scale 0 Scale 1 Scale 2
Ensemble Images
Tumor Region
![Page 18: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/18.jpg)
ResultsOriginal Image Scale 0 Scale 1 Scale 2
Ensemble Images
Tumor Region
![Page 19: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/19.jpg)
Results
![Page 20: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/20.jpg)
Results
Separation at Scale 0 Separation at Scale 1 Separation at Scale 2
![Page 21: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/21.jpg)
CAD Conclusions Using the Hierarchical cascade allows for fast,
accurate analysis of large biomedical images Such a methodology could be employed in a
number of scale-sensitive image analysis systems Both the cascade and the analysis itself can be
applied to various pathologies and imaging modalities
The separation of classes as the scales increase indicates that more discriminatory information is available at higher scales
Higher scales will require development of other feature ensembles to define different Gleason patterns
![Page 22: Prostate Cancer CAD Michael Feldman, MD, PhD feldmanm@mail.med.upenn.edu Assistant Professor Pathology University Pennsylvania](https://reader036.vdocuments.net/reader036/viewer/2022081514/56649f3d5503460f94c5d4b7/html5/thumbnails/22.jpg)
Collaborators
CAD Anant Madhabushi
Scott Doyle