Interdisciplinary Initiatives Program Round 5 – 2010

Daphne Koller, Computer Science
Matt Van de Rijn, Pathology

The goal of this project was to develop and use state-of-the-art techniques from computer vision to create clinically useful tools for the diagnosis and management of breast cancer. To accomplish this goal, we developed an image analysis system to automatically measure thousands of important microscopic characteristics of the size, shape, and behavior of breast cancer cells. We used these measurements to create a tool to aid physicians in the diagnosis of breast cancer and to allow them to more accurately predict patient outcome based on the analysis of the microscopic features of a patient’s breast cancer. We believe that our computational system (based on the quantification of novel quantitative morphologic features) will represent a significant improvement over the current technique of predicting patient outcome based on the manual analysis of a small number of microscopic cancer characteristics, an analysis which is coarse-grained in its resolution, laborious, and subjective.

Building on this work, we propose to analyze both gene expression and microscopic image data from the set of breast cancer patients to better understand how molecular changes in cancer cells produce specific changes in the size, shape, and behavior of cancer cells. This project will allow us to develop clinically useful tools to aid physicians in the diagnosis of invasive breast cancer and will lead to a better understanding of cancer biology, leading to improved breast cancer diagnosis and improved treatments for patients.