A real-time deep learning-based system for colorectal polyp size estimation by white-light endoscopy: development and multicenter prospective validation

Abstract

Background & Aims: Polypectomy devices and surveillance intervals of colorectal polyps are primarily decided by polyp size. We developed a deep learning-based system (ENDOANGEL-CPS) to estimate colorectal polyp size in real time. Methods: ENDOANGEL-CPS calculated polyp size by estimating the distance from endoscopic lens to the polyp and parameters of the lens. The depth estimator network was developed on 7297 images of 5 virtual colon videos and tested on 730 images of 7 virtual colon videos. The performance of the system was first evaluated in nine videos of a simulated colon attached with polyps, then tested in 157 real-world prospective videos from 3 hospitals, comparing with that of operators. It further compared with nine endoscopists in 69 videos. Inappropriate surveillance recommendations caused by wrong estimation of polyp size were also analyzed. Results: The relative error of depth estimation was 11.31%±6.01% in successive virtual colon images. The concordance correlation coefficients (CCC) between system estimation and ground truth were 0.887 and 0.929 in images of a simulated colon and multicenter videos of 157 polyps. The CCC of ENDOANGEL-CPS surpassed all endoscopists (0.890 vs. 0.412±0.290, p<0.0001). The relative accuracy of ENDOANGEL-CPS is significantly higher than that of endoscopists (89.86% vs. 54.74%, p<0.0001). Regarding inappropriate surveillance recommendations, system's error rate is also lower than that of endoscopists (1.45% vs. 16.60%, p<0.0001). Conclusions: ENDOANGEL-CPS could potentially improve the accuracy of colorectal polyp size measurements and size-based surveillance intervals.