Is Artifical Intelligence (CAD EYE) Useful to Not Only Detect But Also to Characterize Small Colorectal Polyps? First Results From a Prospective French Multicenter Study

Aims ”Resect and discard” strategies are difficult to implement since optical diagnosis is challenging. Our aim was to assess the feasibility and performances of a new commercially available system for colorectal polyps.

Methods Colonoscopies were performed in 3 centers by 6 expert endoscopists using 700-series colonoscopes equipped with AI (CAD EYE, Fujifilm). Firstly, AI was activated in the caecum in detection mode. Secondly, when a lesion was detected, the endoscopist was asked to swith AI “off “ and to make a prediction of histology [adenoma, sessile serrated lesion (SSL), hyperplastic polyp (HP)] using the Blue Light Imaging (BLI) mode. This prediction was reported on a dedicated CRF. Thirdly, AI was activated again in BLI mode (“Neoplastic” referring to adenomatous lesions and “Hyperplastic” to HP). All predictions were compared to final histology. A simplified analysis was performed grouping HP and SSL as ‘hyperplastic lesions’ in order to compare CAD EYE and endoscopists’ diagnostic performances.

Results Overall, 153 polyps in 88 patients were retrieved endoscopically and fully documented by histology (n=86 adenomas, n=56 HP, n=11 SSL). Polyp was detected by CAD EYE before the endoscopist in 56/156 (36 %) of cases. Mean polyp size was 4.9 mm [range 1-10 mm]. Polyp morphology was Paris Ip (4 %), Is (18 %), IIa (71 %), IIb (7 %). Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of CAD EYE were 82 %, 94 %, 95 % and 81 %, respectively vs 88 %, 91 %, 93 % and 86 % for expert endoscopists. Diagnostic accuracy of CAD EYE and expert endoscopist were 88 % and 90 %, respectively (p = NS). While CAD EYE classified all SSL as “hyperplastic”, endoscopists were able to classify correctly 8/11 of these lesions as SSL.

Conclusions CAD EYE system shows promising results not only for detection but also for characterization of small colorectal polyps <10mm. However, these results need to be confirmed in larger studies.

Citation: Coron E, Vanbiervliet G, Prouvost V et al. OP119 IS ARTIFICAL INTELLIGENCE (CAD EYE) USEFUL TO NOT ONLY DETECT BUT ALSO TO CHARACTERIZE SMALL COLORECTAL POLYPS? FIRST RESULTS FROM A PROSPECTIVE FRENCH MULTICENTER STUDY. Endoscopy 2021; 53: S50.

Publication History

Article published online:
19 March 2021