Artificial intelligence (AI) has increasingly been employed in multiple fields, and there has been significant interest in its use within gastrointestinal endoscopy. Computer-aided detection (CAD) can potentially improve polyp detection rates and decrease miss rates in colonoscopy. However, few clinical studies have evaluated real-time CAD during colonoscopy. In this study, we analyze the efficacy of a novel real-time CAD system during colonoscopy. This was a single-arm prospective study of patients undergoing colonoscopy with a real-time CAD system. This AI-based system had previously been trained using manually labeled colonoscopy videos to help detect neoplastic polyps (adenomas and serrated polyps). In this pilot study, 300 patients at two centers underwent elective colonoscopy with the CAD system. These results were compared to 300 historical controls consisting of consecutive colonoscopies performed by the participating endoscopists within 12months prior to onset of the study without the aid of CAD. The primary outcome was the mean number of adenomas per colonoscopy. Use of real-time CAD trended towards increased adenoma detection (1.35 vs 1.07, p=0.099) per colonoscopy though this did not achieve statistical significance. Compared to historical controls, use of CAD demonstrated a trend towards increased identification of serrated polyps (0.15 vs 0.07) and all neoplastic (adenomatous and serrated) polyps (1.50 vs 1.14) per procedure. There were significantly more non-neoplastic polyps detected with CAD (1.08 vs 0.57, p<0.0001). There was no difference in=10mm polyps identified between the two groups. A real-time CAD system can increase detection of adenomas and serrated polyps during colonoscopy in comparison to historical controls without CAD, though this was not statistically significant. As this pilot study is underpowered, given the findings we recommend pursuing a larger randomized controlled trial to further evaluate the benefits of CAD.
View details for DOI 10.1038/s41598-022-10597-y
View details for PubMedID 35449442