A novel cloud-based artificial intelligence for real-time detection of colorectal neoplasia - a randomized controlled trial (EAGLE).

Previously, colorectal polyp computer-aided detection (CADe) systems required on-site high-performance hardware installations (e.g., FPGAs/GPUs), creating practical challenges to upgrades and tying hospitals to legacy hardware. Cloud-based CADe solutions overcome these constraints. Hospitals can use low-specification/low-cost hardware to stream data to the cloud for analysis, enabling frequent AI hardware and algorithm updates. Furthermore, existing CADe systems' benefits are largely limited to smaller, less clinically relevant polyps ( < 10 mm). This parallel-group RCT evaluated a real-time cloud-deployed CADe-system trained on an enhanced dataset of clinically significant polyps (large polyps( ≥ 10 mm) and sessile-serrated-lesions(SSLs)). Patients from eight centers across four European countries (841 patients, 22 endoscopists) were randomized to standard or CADe-assisted colonoscopy. Co-primary endpoints were (1) superior Adenomas Per-Colonoscopy (APC), (2) non-inferior Positive Percent-Agreement (PPA) (proportion of resections confirmed as clinically relevant polyps). CADe improved (p < 0.05): APC (0.82 vs. 0.62, Ratio 1.33[95% CI 1.06-1.67]), adenoma detection-rate (43.2% vs. 35.9%), SSL (0.08 vs. 0.03, Ratio 3.30[95% CI 1.41-7.57]), and large polyp (0.12 vs. 0.05, Ratio 2.36[95% CI 1.33-4.17]) detection. PPA was non-inferior, and average cloud-network latency was 59.4 ms per minute, with 99.6% under the 100 ms threshold required for real-time use. This RCT demonstrates the feasibility and efficacy of a real-time cloud-based CADe system, with promising outcomes for clinically significant polyps (large polyps and SSLs). Future research should explore optimizing CADe systems' performance. ClinicalTrials.gov (NCT05730192[15/02/2023]).