New AI Can Tell Normal From Abnormal Colon Biopsies, Would Reduce Pathology Workload
Douglas K. Rex, MD, MASGE, reviewing Graham S, et al. Gut 2023 May 12.
In some countries, pathologists and pathology departments are overworked. A number of artificial intelligence (AI) programs have shown substantial accuracy in interpretation of whole slide images.
About one-third of endoscopic colon biopsies are reported as normal. In this study, an AI model was trained to differentiate normal histologic pictures of colon mucosa from samples showing any abnormality, including neoplasia or inflammation.
At a sensitivity threshold of 99%, the model was shown to reduce the number of normal slides requiring review by a pathologist by 54% in groups of slides from 3 centers used for external validation. At the same sensitivity, the total fraction of large-bowel biopsy slides that could be reduced (not require pathologist review) at 4 UK centers would be 32%, 31%, 17%, and 13%.
Note to readers: At the time we reviewed this paper, its publisher noted that it was not in final form and that subsequent changes might be made.
Graham S, Minhas F, Bilal M, et al. Screening of normal endoscopic large bowel biopsies with interpretable graft learning: a retrospective study. Gut 2023 May 12. (Epub ahead of print) (https://doi.org/10.1136/gutjnl-2023-329512)