EndoScreener

The EndoScreener is a computer-assisted detection software employing a customized deep learning model to provide real-time identification of potential colorectal polyps during white light colonoscopy. Clinical and nonclinical studies demonstrate its ability to enhance polyp detection, notably by reducing adenoma miss rates and increasing adenomas found per colonoscopy, thereby aiding physicians in improving procedural outcomes.

EndoScreener is a stand-alone software designed for the real-time automatic detection of colorectal polyps during colonoscopy procedures in adult patients, utilizing white light colonoscopy. It employs a customized deep learning model, an artificial intelligence algorithm, to analyze live video streams from endoscopy devices. The system assists licensed endoscopists by highlighting potential polyps with blue boxes overlaid on the colonoscopy image and offers an optional sound alert when a polyp is detected; however, physicians retain full responsibility for reviewing these suggestions, confirming the presence or absence of polyps based on their medical judgment, and making any diagnostic decisions, as the software is not intended to replace a full patient evaluation or confirm a diagnosis.\n\nNonclinical testing of EndoScreener, which included validation of its deep learning algorithm on a dataset of 1,138 consecutive polyp patients with histology confirmation, demonstrated acceptable performance regarding per-image sensitivity and specificity, as well as per-polyp sensitivity and Area Under Curve (AUC), with the polyp detection accuracy meeting expectations. Furthermore, the software produced no imaging degradation and ignorable end-to-end latency when used with appropriate hardware. Clinical performance was evaluated in a multi-center, tandem colonoscopy, randomized controlled trial involving 223 patients across four U.S. academic medical centers. This study revealed that initiating the procedure with EndoScreener (CADe-first group) led to a significantly lower adenoma miss rate (AMR) and a higher 1st pass adenoma per colonoscopy (APC) compared to the group that started with routine colonoscopy.

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. What specific architectural details of the "customized deep learning model" differentiate EndoScreener from other AI polyp detection systems, and how do these differences impact performance on various polyp morphologies or sizes?; How does the EndoScreener's performance vary with different types of colonoscopy equipment (e.g., different manufacturers, resolutions) beyond the "appropriate hardware components" mentioned?; What was the quantitative improvement in adenoma miss rate (AMR) and adenoma per colonoscopy (APC) observed in the clinical trial, and were there any specific patient subgroups or polyp characteristics where the benefit was more or less pronounced?; Are there plans to evaluate the long-term impact of EndoScreener use on colorectal cancer incidence or patient outcomes beyond immediate polyp detection rates?; What training is recommended or required for endoscopists to effectively integrate EndoScreener into their workflow and interpret its outputs without over-reliance or alert fatigue?