Computer-aided diagnosis of colorectal polyps: assisted or autonomous?

Abstract

Computer-aided diagnosis (CADx) in colonoscopy aims to improve the accuracy of diagnosing small polyps; however, its integration into clinical practice remains challenging. Human-artificial intelligence (AI) collaboration, which is expected to enhance optical diagnosis, has shown limited success in clinical trials, with studies indicating no significant improvement in human-only performance. Conversely, autonomous CADx systems that operate independently of clinicians have demonstrated superior diagnostic accuracy in some studies, suggesting their potential for efficiency, consistency, and standardization in healthcare. However, the adoption of autonomous AI raises ethical, legal, and practical concerns such as accountability for errors, loss of clinical context, and clinician or patient distrust. The decision between using CADx as an assistant or as an autonomous system may depend on the clinical scenario. Autonomous systems can standardize routine screening for low-risk patients, whereas assistive systems may complement expertise in complex cases. Regardless of the model used, robust regulatory frameworks and clinician training are essential to ensure safety and maintain trust. Balancing the strengths of AI with the critical role of human judgment is the key to optimizing outcomes and navigating the complex implications of integrating CADx technologies into colonoscopy practice.