Machine Learning-Based Quantitative Evaluation of Histological Disease Severity in Ulcerative Colitis

Emerging evidence supports the adoption of histologic improvement as a therapeutic goal and clinical trial endpoint for patients with ulcerative colitis (UC), including reports that show relapse has been associated with residual histologic disease activity in patients who achieved endoscopic endpoints.

Here we show sensitive and reproducible machine learning (ML)-based pixel-level identification and quantification of histological features of UC, correlation with expert manual pathologist scoring of disease severity, and use of model-derived histological features for prediction of UC disease severity scores.