Machine learning in the differential diagnosis of ulcerative colitis and Crohn's disease: a systematic review.

Background: Inflammatory bowel disease (IBD) is a complex chronic disease of the gastrointestinal tract. This systematic review aimed at highlighting the latest findings on the use of machine learning (ML) in the IBD subtypes, ulcerative colitis and Crohn's disease (CD), with a view to obtaining a basis for the clinical application of ML to differentiate between these subtypes. Methods: We conducted an extensive search of six major databases, including PubMed, Web of Science, Embase, Cochrane Library, Scopus, and Ovid, for entries made between 1 January 2000 and 28 November 2024. The search was focused on identifying studies that used ML to construct diagnostic models for ulcerative colitis and CD. Quality Assessment of Diagnostic Accuracy Studies was used to assess the risk of bias and concerns about the applicability of these studies. The protocol for this review was registered in PROSPERO (CRD42024543036). Results: After a rigorous screening and assessment process, 31 papers were found to be suitable for inclusion in the review, with a total sample size of 15,140. Most of the included studies were retrospective (n=27, 87%), with the vast majority of studies (n=20, 65%) published between 2021 and 2023. Random forest (RF) was identified as the most commonly used (n=10, 32%), followed by support vector machines (n=9, 29%), and the majority of the studies were focused on model evaluation metrics of ML. Conclusions: Our findings indicate that ML holds the potential to enhance diagnostic accuracy in distinguishing between ulcerative colitis and CD, particularly through the utilization of models developed from endoscopic and fecal biomarker data based on deep learning and RF.