Breast cancer is a heterogeneous disease and its molecular classification using PAM50 gene expression profiles assumes homogeneity of gene expression across patients that are assigned to a particular sub-type. We propose metrics to quantify this heterogeneity and show differences in clinical and survival characteristics of homogeneous vs. heterogeneous Luminal A breast cancer patients. We also show that detection of geographic foci of individual sub-clonal mutations within the same tumor slide may be possible using pattern analysis of H&E stained slides with convolutional neural networks. H&E stained slides are much cheaper to prepare than getting a PAM50 gene profile. We hope that linking our predictive analysis with treatment effectiveness will provide valuable insights into tumor biology and will contribute to personalized treatment of cancers.
Amit Sethi is an Associate Professor of Electrical Engineering at Indian Institute of Technology Bombay and an Adjunct of Pathology at University of Illinois at Chicago. His research interests lie in machine learning and deep learning and their applications to images and videos particularly in medical imaging. He obtained his post-secondary education at Indian Institute of Technology Delhi and University of Illinois at Urbana-Champaign. He has previously worked at NEC Labs America in Cupertino CA, ZS Associates in Chicago, and Indian Institute of Technology Guwahati.