About the Article
Article Files
[This article belongs to Volume - 69, Issue - 9]
Published on : 2024-09-05 12:46:13
Article Code: AMJ-05-09-2024-12300
Title : AI-Augmented Digital Pathology Predicts Mismatch Repair Deficiency in Colorectal Cancer with Superior Sensitivity to Immunohistochemistry
Author(s) : Dr. Sarah Okafor, Dr. Li Wei, Dr. Ahmed Hassan
Abstract :
Lynch syndrome screening relies on immunohistochemistry (IHC) for mismatch repair (MMR) proteins, which suffers
from pre-analytic variability. We developed a deep learning algorithm (PathMMR-AI) trained on 5,000 whole-slide
images (WSIs) of colorectal cancer (CRC) to predict MMR deficiency (dMMR) directly from hematoxylin-eosin (H&E)
stained slides. The model achieved 96.8% sensitivity and 94.2% specificity, outperforming conventional IHC
(sensitivity 89.3%) in a prospective validation cohort of 400 cases. Attention mapping revealed morphological
features (tumor-infiltrating lymphocytes, mucinous differentiation, Crohn's-like aggregates) predictive of dMMR
status. AI-augmented H&E screening offers a cost-effective, rapid triage tool for universal Lynch syndrome screening
in resource-limited settings.