Title: Deep learning enables colon adenocarcinoma patient stratification from whole slide images.

Abstract: Colorectal cancer is one of most common cancers in both men and women. While treatments have improved over the last few decades, effective identification of high risk and potentially relapsing patients has been a challenge. Current standard-of-care relies on tumor boards discussing each patient individually, which is time consuming and susceptible to subjectivity amongst different experts. Machine learning and deep learning can systematize patient stratification and may serve tumor boards as a valuable tool for clinical decision making. To that end, we are building deep learning models that flag high risk colorectal cancer patients based on H&E images. Such patients can benefit from more rigorous medical follow up.  Analysis of image regions associated with high risk of relapse can also help pathologists refine clinical decisions. In this talk I will go over the challenges of developing deep learning models for survival analysis using whole slide images and discuss alternative approaches for patient classification. In particular, heterogeneity within slides supports a tile-based analysis approach. Furthermore, deep learning methods based on binary classification enjoy high prediction accuracies in stratifying patients.

Bio of the speaker: Dr. Ali Foroughi pour obtained his BSc in electrical engineering from Sharif University, Iran in 2013, and his MSc in mathematics and PhD in electrical engineering from the Ohio State University in 2019. He then joined the Jackson Laboratory for Genomic Medicine as a postdoctoral researcher in June 2019. Dr. Foroughi pour was awarded Ohio State’s Distinguished University Fellowship in 2013, and the Jackson Laboratory’s JAX Scholar Award in 2020. During his graduate studies Dr. Foroughi pour developed a new Bayesian approach for small-sample high-dimensional feature selection technique tailored for biomarker discovery, and he developed flexible classifier design strategies based on the high dimensional model representation theory for analyzing -omics datasets. As a postdoctoral researcher in JAX he has been working on developing reliable deep learning models for analysis of large scale hematoxylin and eosin (H&E) stained whole slide images, as well as integrative analysis of H&E slides and multiplexed immune mass cytometry (IMC) images. His recent work, published in Nature Communications, depicts the cross-classification capability of convolutional neural networks (CNNs) on tumors.

Registration Link: https://libcal.dartmouth.edu/calendar/itc/2021DSAIW1