EW10-EC02 Endoscopy Support Program

The EW10-EC02 Endoscopy Support Program is an AI-powered software designed to enhance real-time detection of colonic lesions like polyps and adenomas during colonoscopies using both White Light Imaging and Linked Color Imaging. Its methodology leverages a deep learning algorithm to analyze live endoscopic video and highlight suspicious areas, with clinical trials demonstrating its efficacy in increasing the number of adenomas and polyps detected per colonoscopy when used as an aid by endoscopists.

The EW10-EC02 Endoscopy Support Program is an artificial intelligence (AI)-driven software designed as a computer-assisted reading tool for endoscopists. Its primary purpose is to aid in the real-time detection of colonic mucosal lesions, such as polyps and adenomas, during standard colonoscopy examinations for both screening and surveillance populations. The system employs a deep learning-based algorithm that analyzes endoscopic video images obtained through standard White Light Imaging (WLI) and Linked Color Imaging (LCI). This computer-aided diagnosis (CAD) technology functions by superimposing visual markers on suspected lesions identified in the live video feed, thereby intending to enhance the physician's visual assessment capabilities without replacing their ultimate clinical judgment. Non-clinical testing of the EW10-EC02 demonstrated high lesion-based sensitivities (95.1% for WLI, 95.5% for LCI) and acceptable false positive rates per patient, with minimal video processing delay attributed to the software. A subsequent multi-center, prospective, randomized controlled trial involving 1,031 subjects compared computer-assisted colonoscopy (CAC) utilizing the EW10-EC02 with conventional colonoscopy (CC). The results indicated that the CAC group achieved a statistically significant increase in Adenoma per Colonoscopy (APC) (0.990 vs 0.849), Polyp per Colonoscopy (PPC) (1.680 vs 1.328), Serrated Lesions per Colonoscopy (SLPC) (0.690 vs 0.475), and Serrated Lesions Detection Rate (SLDR) (35.6% vs 29.1%). While the pre-defined criteria for Positive Predictive Value (PPV) and Positive Percent Agreement (PPA) were met, the absolute values for the CAC group were slightly lower than those for the CC group. The False Positive Rate (FPR) for CAC was determined to be non-inferior to that of CC. Importantly, no significant differences were observed in the rates of adverse events or in colonoscopy withdrawal times between the two study arms.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the algorithm's performance differ between White Light Imaging (WLI) and Linked Color Imaging (LCI) in terms of specific lesion morphologies (e.g., flat vs. polypoid) and sizes, beyond the overall sensitivity and false positive rates reported in the non-clinical studies?; Given the slightly lower Positive Predictive Value (PPV) and Positive Percent Agreement (PPA) for the Computer Assisted Colonoscopy (CAC) group in the clinical trial, alongside a non-inferior but numerically higher False Positive Rate (FPR), what is the potential impact on unnecessary biopsies or resections, and how might this be mitigated through endoscopist training or algorithm refinement?; The clinical study noted a statistically significant interaction between screening versus surveillance colonoscopies for the Adenoma per Colonoscopy (APC) endpoint. What specific factors inherent to surveillance procedures (e.g., patient history, expected lesion characteristics) might explain the more pronounced benefit of the EW10-EC02 AI tool in these cases compared to initial screening procedures?; The non-clinical performance testing utilized a dataset from a 100% Asian patient population, while the pivotal clinical trial was conducted in the US with a more diverse racial demographic. Were any sub-analyses performed in the clinical trial to assess if the algorithm's performance (sensitivity, specificity, FPR) varied across different racial or ethnic groups, and how does this inform its generalizability?; What were the specific visual characteristics of the false positives most commonly generated by the EW10-EC02 system during the clinical trial (e.g., mucosal folds, debris, vascular patterns, suction artifacts), and how do these compare in nature and frequency to false positives typically encountered by endoscopists performing conventional colonoscopy without AI assistance?