DML Sharif University of Technology
Real-time automatic detection and classification of colorectal polyps during colonoscopy using Explainable AI
Lately, the ASGE1 has addressed the resect and discard strategy known as PIVI2, determining that if a polyp is less than 5mm, we can omit histopathological examinations if we are highly confident with our diagnosis. PIVI also states that hyperplastic polyps in the rectosigmoid colon can be left in place without sampling or endoscopic resection owing to its nonmalignant nature. If we can differentiate polyps, we can reduce unnecessary endoscopic resections, which may, in turn, decreases complications, physician burden, and medical costs. Hence, we aimed to build an AI system that can accurately detect and classify (differentiate) CP(Colorectal Polyp) during colonoscopy in real-time. We want to perform a prospective, multi-center (multi-dataset) study of patients referred for colonoscopy. Procedural videos will be analyzed by a validated deep-learning AI polyp detection software that labels suspected polyps. There are a lot of public annotated datasets for polyp detection in colonoscopy.
We will utilize them to train a sophisticated multi-center AI system. Next, we will tune the model by a national dataset of static images captured and labeled by experienced endoscopists during colonoscopy. Some of the most referred databases such as CVC-300, CVC-612, Kvasir, Kvasir-SEG are entirely downloaded. We are currently working to implement SOTA methods such as YOLOR, YOLOX, etc., to our datasets.
Start Date
Sept. 22, 2020
Amir Pourmand
Rassa Ghavami
Hamid R. Rabiee