Impact of User’s Background Knowledge and Characteristics of Colonic Polyps on Lesion Recognition during Colonoscopy with Computer-aided Detection

Abstract

Abstract Interaction between endoscopists and computer-aided detection (CADe) could be crucial in determining the effectiveness of CADe-assisted colonoscopy. This study investigated the effects of CADe on the recognition of diverse colorectal polyps by the endoscopy department staffs with varying experience levels. A computerized test module with 300 colonoscopy images was developed to measure changes in the polyp recognition performance with or without CADe assistance. The effect sizes of CADe for the nurse, fellow, and expert groups were evaluated based on polyp features including histopathology and detection difficulty. The CADe system demonstrated the following standalone performance rates during polyp detection: 79.0% accuracy, 78.5% sensitivity, and 80.3% specificity. Detection accuracy among participants was significantly improved with CADe assistance (odd ratio, 1.88; p < 0.001). Furthermore, it was observed that when the CADe system was precise, the likelihood of participants accurately identifying lesions increased by an average of 2.87fold (odd ratio, nurse group: 6.78; fellow group: 2.15; expert group: 2.18). However, synergistic effect that exceeded the standalone performance of CADe was only observed for the detection of adenomas and easily detectable lesions. The effect size and synergism of CADe and humans for polyp recognition could vary based on the user’s experience level and polyp characteristics.