GI Genius

The GI Genius system is an artificial intelligence tool that assists endoscopists by processing real-time colonoscopy video to highlight potential colonic mucosal lesions like polyps, thereby aiding in their detection during standard white-light examinations. Its software version 2.0.0 leverages an enhanced neural network, retrained with more extensive data and improved augmentation, alongside an optimized inference engine, to achieve superior detection performance, speed, and energy efficiency compared to earlier versions.

The GI Genius System is an artificial intelligence-based, computer-assisted reading tool designed to support endoscopists in detecting colonic mucosal lesions, such as polyps and adenomas, in real-time during standard white-light endoscopy examinations for screening and surveillance. It integrates between the video processor and endoscopic display, processing live colonoscopy images and highlighting potential lesions with superimposed green square markers and an audible cue. The system is intended to augment, not replace, clinical decision-making and is compatible with specific video processors, including Fujifilm, Olympus, and Pentax models. This document outlines the updates in GI Genius software version 2.0.0 compared to its predecessor (version 1.0.2). Key changes include an upgraded inference engine for enhanced algorithm speed and energy efficiency, and a retrained neural network using additional procedure videos and improved data augmentation, aiming for more robust polyp detection. Non-clinical performance testing demonstrated improvements: lesion-based sensitivity increased from 82.0% to 86.5%, frame-level true positive detection improved, false positive rates per frame decreased (from 2.02% to 1.96%), and the Frame-Based True Positive Rate/False Positive Rate ROC curve Area Under Curve (AOC) improved from 0.723 to 0.796. The updated software also adds compatibility with the Fujifilm VP-7000 video processor.

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 improved polyp detection capability and robustness of software version 2.0.0, particularly the increased lesion-based sensitivity from 82.0% to 86.5%, translate to potential changes in adenoma detection rates (ADR) in diverse real-world clinical populations?; What specific data augmentation techniques were employed in retraining the neural network for version 2.0.0, and how significantly did these contribute to the observed improvements in true positive rates and the reduction in false positive rates per frame compared to version 1.0.2?; Considering the device is limited to standard white-light endoscopy, what are the technical challenges or future development plans for expanding its utility to advanced endoscopic imaging modalities like narrow-band imaging (NBI) or chromoendoscopy?; The document mentions an 'upgrade to the inference engine to optimize algorithm speed and improve energy efficiency.' Can the improvement in energy efficiency be quantified, and what is its practical impact in a clinical setting (e.g., on device temperature, longevity, or power requirements)?; Beyond the per-frame and lesion-based sensitivity metrics, were there any evaluations on the system's performance regarding the detection of specific challenging lesion morphologies, such as flat or sessile serrated adenomas, in the new software version?