Real-time Diagnosis of Diminutive Colorectal Polyps Using AI

Correct endoscopic prediction of the histopathology and differentiation between benign, pre-malignant, and malignant colorectal polyps (optical diagnosis) remains difficult. Artificial intelligence has great potential in image analysis in gastrointestinal endoscopy. Aim of this study is to investigate the real-time diagnostic performance of AI4CRP for the classification of diminutive colorectal polyps, and to compare it with the real-time diagnostic performance of commercially available CADx systems.

Correct endoscopic prediction of the histopathology and differentiation between benign, pre-malignant, and malignant colorectal polyps (optical diagnosis) remains difficult. Despite additional training, even experienced endoscopists continue to fail meeting international thresholds set for safe implementation of treatment strategies based on optical diagnosis. Multiple machine learning techniques - computer-aided diagnosis (CADx) systems - have been developed for applications in medical imaging within colonoscopy and can improve endoscopic classification of colorectal polyps. Aim of this study is to explore the feasibility of the workflow using AI4CRP (a CNN based CADx system) real-time in the endoscopy suite, and to investigate the real-time diagnostic performance of AI4CRP for the diagnosis of diminutive (\<5mm) colorectal polyps. Secondary, the real-time performance of commercially available CADx systems will be investigated and compared with AI4CRP performance.