Artificial intelligence for deep profiling and characterization of colorectal polyps

Colorectal cancer (CRC) is a major cause of cancer deaths. It develops from precancerous lesions, called polyps, which can be endoscopically detected and removed. Recent insights in the basic molecular and histopathological features of the different types of polyps (hyperplastic polyp - HP, Adenoma - AD and Sessile serrated polyp - SSP) indicate that the classification of polyps and their associated risks differs significantly from the current classical adenoma-carcinoma sequence. Whereas ADs originate from WNT-dependent stem cell expansion, SSPs originate from a different neoplastic pathway, namely metaplasia, in which differentiated cells transdifferentiate into other, non-native celltypes. In SSP a pyloric-type gastric metaplasia has been observed, in which alien MUC5A-positive goblet cells are formed and colonize the colonic crypts from the colonic lumen down to the cryptbase. This fundamentally differs from ADs where stemcells from the crypt base expand towards the colonic lumen. Since metaplasia-based carcinogenesis is a top-down process, we hypothesize that artificial intelligence has the potential to help us discover the correlation between endoscopic image features and the underlying molecular/histopathological features. This may empower a more accurate characterization and risk stratification in real-time and will help further unravel the pathways involved in colorectal carcinogenesis, leading to new strategies in prevention, surveillance, and therapeutics.