DNA mismatch repair deficiency (MMR) impacts approximately 15% of all colorectal cancer patients. Identification of positive or negative MMR status in patients impacts their treatment.

In collaboration with  ProPath and Hologic, we have applied an attention-based multi-instance learning model to accurately predict DNA mismatch repair deficiency from H&E stained colon biopsies. This model could serve as an important decision support tool for pathologists assessing colon cancer.