Role of CADx in colonoscopy: lessons from real-life studies

Screening colonoscopy has significantly reduced colorectal cancer (CRC) incidence and mortality rates. However, its effectiveness relies on the accurate detection and characterization of precursor lesions. Computer-aided diagnosis (CADx) systems leveraging artificial intelligence (AI) have emerged as potential tools to enhance lesion classification and optimize clinical decision-making. This review critically examines the real-world performance of CADx in colonoscopy, highlighting its benefits and limitations in lesion characterization. While early studies demonstrated high sensitivity and negative predictive value (NPV), real-life clinical implementation has shown inconsistencies due to variability in endoscopist experience, lesion morphology, and AI training datasets. Additionally, challenges such as algorithm reliability, human-AI interaction, and integration into clinical workflows persist. This article explores the evolving role of CADx in supporting optical diagnosis, its impact on resect-and-discard strategies, and the future directions required to maximize its clinical utility.