Computer-aided detection (CADe) and diagnosis (CADx) systems aim to improve colonoscopy quality. This study evaluated the real-time performance of the CAD EYE system (Fujifilm, Japan) for both polyp detection and the optical characterization of colorectal lesions into hyperplastic vs. neoplastic categories.

A prospective, comparative study was conducted. Consecutive patients undergoing colonoscopy were allocated to procedures either with (study group, n=56) or without (control group, n=54) CAD EYE assistance. The system was used in CADe mode with white light imaging (WLI) and linked color imaging (LCI). For characterization (CADx), blue light imaging (BLI) mode was used to predict histology (neoplastic vs. hyperplastic). Primary outcomes were the sensitivity and specificity of CADe for polyp detection and of CADx for optical diagnosis. Secondary endpoints included procedural metrics. Statistical analysis included calculation of performance metrics with 95% confidence intervals and AUC.

A total of 241 lesions (mean size 7.9 mm) were analyzed from 110 patients: 135 (56%) neoplastic and 106 (44%) hyperplastic. Patient groups were well-matched for bowel preparation (Boston Bowel Prep Score >7: 87.5% vs 87%) and cecal intubation (100% each). The CADe system demonstrated a sensitivity of 97%, specificity of 84%, accuracy of 93%, and an AUC of 0.95 for polyp detection. For optical characterization, the CADx mode achieved a sensitivity of 96%, specificity of 99%, accuracy of 98%, and an AUC of 0.96 for differentiating neoplastic from hyperplastic lesions.

The CAD EYE system shows high accuracy for both real-time polyp detection and optical characterization of colorectal lesions, performing at expert-level benchmarks. Its integration into clinical practice has the potential to enhance the quality of screening colonoscopy, increase adenoma detection rates (ADR), and support real-time endoscopic decision-making. Prospective multicenter studies are warranted to confirm its impact on long-term patient outcomes.