Xiao Xiao, PhD, JingJia Liu. Shanghai Wision AI Co., Ltd.
Realtime automatic polyp detection during colonoscopy reduces the number of missed colon polyp/adenoma in the visual field, which help to improve adenoma detection and thus decrease the risk of interval colorectal cancer.?
Real-time automatic polyp-detection system detects each colon polyp with a blue hollow box on an adjacent monitor during colonoscopy. It serves as an effective second observer which draws the endoscopist’s attention, in real time, to concerning lesions, effectively creating an ‘extra set of eyes’ on all aspects of the video data with fidelity. The key competence of the system is the high sensitivity and specificity detection performance, which is obtained by training a deep learning network with 5545 annotated colonoscopy images.
In a preclinical validation, 27113 colonoscopy report images from 1138 consecutive patients were used as validation dataset, and the system achieved a per-image-sensitivity of 94.38% and a per-image-specificity of 95.52%. Tested on a video series of 138 consecutively encountered polyps from 151 patients who underwent colonoscopy examination, the per-image-sensitivity was 91.64% and per-lesion-sensitivity 100%. Tested on 54 unaltered full-length colonoscopy videos with no polyp, the per-image-specificity was 95.40%. With an NVIDIA Titan X pascal GPU, the system detects 30 frames per second with a detection latency of 76ms. The full study has been published by Nature Biomedical Engineering Issue 10, Vol 2, 2018.
In a clinical trial on Chinese population, consecutive patients were prospectively randomized to undergo routine colonoscopy with or without assistance of a real-time automatic polyp detection system providing a simultaneous visual notice and sound alarm when a polyp was detected. Out of 1,058 patients, 536 were randomized to standard colonoscopy, and 522 to colonoscopy with computer-aided diagnosis.A total of 767 polyps, 422 adenomas and 31 serrated adenomas were detected. The ADR were 20.34% and 29.12% (P< 0.001) of the control and CAD groups respectively. The average number of adenomas were 0.31 and 0.53 (P< 0.001). The study has been presented on United European Gastroenterology Week 2018 and received National Scholar Award.
The system increases the number of adenomas found per colonoscopy and overall ADR significantly in low prevalent ADR population, more trials have been started in different regions in the world. Beth Israel Deaconess Medical Center of Harvard Medical School and NYU Langone Health are the two primary centers of clinical validation on US population.
Presented at the SAGES 2017 Annual Meeting in Houston, TX.
Abstract ID: 98847
Program Number: ETP762
Presentation Session: Emerging Technology Poster Session (Non CME)
Presentation Type: Poster