Refining risk stratification

New approach to refining risk stratification for colorectal cancer patients

This study seeks to harness deep learning methods to identify unique patterns across colorectal cancer (CRC) tumor slides for improved disease classification using 300,000 images collected already from over 6,000 patients.

This will provide a novel addition to classic histopathology and molecular methods for improved treatment and management of CRC.

Link to Pathology University Bern

Lay Abstract

Colorectal cancer is the third leading cause of cancer-related mortality in Switzerland and despite improvements in therapy for patients with metastatic disease, the overall 5-year survival rate remains low, at 60%. Biomarkers for risk-stratification in patients with non-metastatic disease remain scarce. The tumor microenvironment, however, can provide a wealth of potential new biomarkers as it consists of a rich network of different types, frequencies and densities of cell types, which may also reflect the underlying molecular aberrations of the cancer. These interactions are complex and can only be comprehensively analyzed using computational methods. In this study, we use the histopathology images of colorectal cancer tissues to study these cellular interactions using artificial intelligence approaches, including deep learning. We identify cancers with particular molecular defects based on patterns within the tissues and study the impact of these patterns on patient outcome.