P1978 - Diagnostic Accuracy of Artificial Intelligence in Classification of Colonic Polyps – A Real World Prospective Study

Rajkumar P. Wadhwa, MD, DM, Aathira Ravindranath, , Kushan Sengupta, , Shantanu Shahane, , Nenavath Raghuram Naik,
Apollo BGS Hospitals, Mysore, Karnataka, India

Introduction: Colonoscopy is used to detect colonic polyps early and in turn to prevent colonic cancer related deaths. Apart from increasing the rate of detection of colonic polyps using artificial intelligence, classification as neoplastic and non-neoplastic can improve the endoscopic diagnostic capabilities. The aim of this study was to assess the utility of CAD eye (Fujifilm, Tokyo, Japan) in characterization of colonic polyps.

Methods: This was a single-center prospective study conducted from June 2022 to April 2023. Polyps detected during routine colonoscopy were characterized as Neoplastic or Hyperplastic using CADeye technology with magnified blue-light imaging (BLI). The size of the polyps and the CADeye classification were noted. All polyps were sent for histopathological examination either by biopsy or resection and they were examined by a pathologist who was blinded to the CADeye classification.

Results: Among 56 patients 106 polyps were detected. Median size of the polyp was 5mm (range: 2 to 10). Of them 61 were characterized as neoplastic and 45 as hyperplastic under CADeye with magnified BLI. Of the 61 polyps that were characterized as neoplastic, histopathology showed villous adenoma (n=24), tubulovillous (n=20), malignant (n=12), tubular adenoma (n=3) and hyperplastic (n=2). Among 45 polyps that were hyperplastic on CADeye histopathology showed hyperplastic polyp (n=22), inflammatory polyp (n=4), juvenile polyp (n=5), tubular adenoma (n=7), tubulovillous adenoma (n=4), villous adenoma (n=1) and sessile serrated adenoma (n=2). For detection of neoplastic polyps sensitivity was 78.67% (67.7 – 87.3), specificity was 93.5% (78.6 – 99.2), positive predictive value 96.7% (88.6 – 99.6), negative predictive value 64.4% (48.8 – 78.1). The overall accuracy was 83.0% (74.5 – 89.6).

Discussion: CADeye can accurately classify polyps as neoplastic in 93.5% cases. CADeye is a useful tool in clinical practice to classify polyps as neoplastic and hyperplastic.


Disclosures:


Rajkumar P. Wadhwa, MD, DM, Aathira Ravindranath, , Kushan Sengupta, , Shantanu Shahane, , Nenavath Raghuram Naik, . P1978 - Diagnostic Accuracy of Artificial Intelligence in Classification of Colonic Polyps – A Real World Prospective Study, ACG 2023 Annual Scientific Meeting Abstracts. Vancouver, BC, Canada: American College of Gastroenterology.