Computer Vision Methods in Surgical Pathology: Diagnosing Carcinoma of the Breast
Alesia Kaplan, Xinyan Li, Ravishankar Sivalingam, Guruprasad Somasundaram, Arindam Banerjee, Vassilios Morellas, Nikolaos Papanikolopoulos, Alexander Truskinovsky. University of Minnesota, Minneapolis, MN
Background: Based on our successful application of computer vision methods to detect endometrial and prostatic adenocarcinomas, we extend our analysis to ductal carcinoma of the breast, using region covariance descriptors.
Design: We used 5 images each of H&E-stained sections of 2 diagnostic classes, carcinoma and benign breast, scanned at x50 magnification on a digital slide scanner. The color images were transformed to grayscale using a custom transformation and then manually annotated to train the classification algorithm. For this study, we have developed a new Graphical User Interface (GUI) to enable a pathologist to annotate tissue images for computer analysis. The annotated regions are then broken down into overlapping blocks of 150 x 150 pixels. Each block is represented by the covariance matrix over the image features in that block. We used a set of spatial (x, y, ρ, θ) and intensity [I, Ix, Iy, √(Ix2+Iy2)] features, giving rise to 8x8 region covariance descriptors (RCDs). Test image patches are classified via k-NN classification using a geodesic distance measure over RCDs. We ran a 10-fold cross-validation on the entire dataset to test performance on unseen images, where we randomly select one tenth as test images from each class, using the rest for training.
Results: Using RCDs with more than 6000 labeled training patches, the overall classification accuracy of the image blocks into cancerous vs. benign regions was 98%. Figure 1 shows a ductal carcinoma on the left, and the same tumor successfully identified and highlighted by the algorithm on the right.
Conclusions: We have demonstrated that computer vision methods can be applied to diagnose ductal carcinoma of the breast with high accuracy. The approach based on RCDs has been previously shown by us to be accurate for endometrial and prostatic adenocarcinomas. The continued success on carcinoma of the breast indicates that RCDs are indeed robust as features for cancer diagnosis. This study also highlights the successful use of our new image annotation GUI. It is simple and intuitive and does not require advanced computer knowledge to operate. Using this GUI, the computer vision algorithms can be trained for new tasks and applied to more and more fields of diagnostic histopathology.
Monday, March 4, 2013 9:30 AM
Poster Session I Stowell-Orbison/Surgical Pathology/Autopsy Awards Poster Session # 250, Monday Morning