ABSTRACT
Background: Artificial intelligence-based optical diagnosis systems (CADx) have been developed to allow pathology prediction of colorectal polyps during colonoscopies. However, CADx systems have not yet been validated for autonomous performance. Therefore, we conducted a trial comparing Autonomous AI (AI-A) to AI assisted human (AI-H) optical diagnosis.
Methods: We performed a randomized non-inferiority trial of patients undergoing elective colonoscopies in one academic institution. Patients were randomized int: 1) AI-A-based CADx optical diagnosis of diminutive polyps without human input; 2) endoscopists performed optical diagnosis of diminutive polyps after seeing the real-time CADx diagnosis. Primary outcome was accuracy in optical diagnosis in both arms using pathology as gold standard. Secondary outcomes included agreement with pathology for surveillance intervals.
Results: 467 patients were randomized (238 patients/158 polyps in the AI-A group; 229 patients/179 polyps in the AI-H group). Accuracy for optical diagnosis was 77.2% (95%Confidence Interval [CI] 69.7-84.7) in the AI-A group and 72.1% (95%CI 65.5-78.6) in the AI-H group (p=0.86). Sensitivity, specificity, PPV and NPV for adenoma diagnosis were 84.8%, 64.4%, 85.6%, and 63.0%, respectively in the AI-A group vs 83.6%, 63.8%, 78.6%, and 71.0% in the AI-H group. Rectosigmoid-NPV was >90% in both groups. AI-A had statistically significantly higher agreement with pathology-based surveillance intervals compared to AI-H (91.5% [95%CI 86.9-96.1] vs 82.1% [95%CI 76.5-87.7]; p=0.016).
Conclusions: Autonomous AI-based optical diagnosis exhibits non-inferior accuracy to endoscopist-based diagnosis but achieves higher agreement with pathology-based surveillance intervals and reached the required quality threshold for Resect-and-discard and Diagnose-and-leave. These strategies can therefore be considered when using current CADx systems autonomously.

Jeremy Liu Chen Kiow

Appearances