Multi-view Framework for Histomorphologic Classification

Stephanie A Harmon, Samira Masoudi, Thomas Sanford, Sherif Mehralivand, Stephanie Walker, Peter Choyke, Brad Wood, Peter Pinto, Jesse McKenney, Baris Turkbey

Abstract: Current routine histopathologic evaluation of prostate cancer does not fully account for some individual morphology patterns associated with poor outcome. Pathologists evaluate and score morphology across multiple magnifications, motivating deep learning methods to incorporate various resolutions. We have evaluated a proof-of-concept multi-view framework to classify high risk morphology architectures that does not rely on ensemble-based techniques of multi-magnification models.

Paper Type: both

TL;DR: generation of multi-view image stacks outperforms individual magnifications in digital pathology classification

Track: short paper

Keywords: Classification, Prostate Cancer, Digital Pathology