大腸ポリープ検出における白色光画像と人工知能支援白色光画像：ランダム化試験

Adenoma detection rate (ADR) and sessile serrated lesion (SSL) detection rate (SDR) are crucial quality indicators for colonoscopy, as their improvement contributes to effective prevention of colorectal cancer. Artificial intelligence (AI) has been shown to significantly increase ADR. This study compared white light imaging (WLI) versus AI-assisted WLI for neoplasia detection.

This was a prospective, randomised trial of screening, surveillance, and symptomatic patients. Our primary objective was to evaluate ADR. Secondary objectives included SDR, mean number of adenomas per patient (MAP), neoplasia detection rate (NDR), advanced ADR (AADR), and colonoscope withdrawal time.

A total of 621 adenomas were diagnosed in 711 patients, with 310 adenomas in the WLI group and 311 adenomas in the WLI+AI group (p=0.65). Eighty-three SSLs and two intramucosal carcinomas were also detected, totalling 706 neoplasms. ADR was 45.9% in the WLI group and 50.8% in the WLI+AI group (p=0.20). ADR was 54.4% for screening, 49.0% for surveillance, and 40.0% for symptomatic patients (p=0.01). Marginal significance was observed in the WLI+AI group for screening patients (61.5% vs. 49.2%, p=0.06). SDR was 9.0% for both groups. MAP (0.9 vs. 0.9, p=0.34), NDR (51.0% vs. 56.8%, p=0.13), and AADR (8.4% vs. 7.6%, p=0.78) did not differ significantly between the groups. Withdrawal time was similar for the WLI (12.4 ± 5.1 min) and WLI+AI (12.2 ± 4.1 min) groups (p=0.32).

AI-assisted colonoscopy demonstrated high ADR and NDR. While without statistical relevance overall, marginal significance was observed for screening patients.