Topologically-Guided Deep Learning for Computational Pathology

 Shahira Abousamra, PhD
Postdoctoral Scholar in the Plevritis Lab, Department of Biomedical Data Science.
 

Date: Sept 19, 2025
11:00am - 12:00pm  
James H. Clark Center, Room S360, 3rd floor next to the Coffee Shop
Zoom link | Meeting ID: 965 6569 9018 | Password: 298181

Abstract

Within the tumor microenvironment, the arrangement and spatial co-localization of various cell types is increasingly being recognized as critical factors influencing cancer progression and drug response. By just looking at pathology images, humans can recognize topological structures formed by cells in various tissue in the form of loops, tunnels, and clusters. However, modern machine learning frameworks seldom explicitly integrate these visually apparent spatial context features in their learning. In this talk I will demonstrate the value of these features by leveraging topological data analysis and spatial statistics.  I will describe the representation of these structural patterns formed by different cell types at the patch level of whole slide images. Leveraging these spatial descriptors in innovative ways enables us to explicitly integrate the spatial context to advance deep learning models for various applications in computational pathology. I will discuss this in the context of modeling the cell spatial context, conditional generation, and model optimization.

About

I am is a Postdoc at the Plevritis lab in the department of Biomedical Data Science at Stanford University. I earned my PhD in Computer Science from Stony Brook University in 2024 under the supervision of Dr. Chao Chen and Dr. Dimitris Samaras. My research spans computer vision, biomedical image analysis, and topological data analysis. I am particulary interested in integrating mathematical modeling with computer vision to create more robust solutions, especially in the context of advancing cancer research and enhancing our understanding of the tumor microenvironment.