Application of deep learning for autonomous detection and localization of colorectal polyps in wireless colon capsule endoscopy. (January 2020)

• Record Type: Journal Article
• Authors: Nadimi, Esmaeil S.; Buijs, Maria M.; Herp, Jurgen; Kroijer, Rasmus; Kobaek-Larsen, Morten; Nielsen, Emilie; Pedersen, Claus D.; Blanes-Vidal, Victoria; Baatrup, Gunnar
• Is Part Of: Computers & electrical engineering. Volume 81(2020)
• Publication Date: 2020-01
• Subjects: Colorectal polyps -- Colon capsule endoscopy (CCE) -- Deep learning -- Machine learning -- Convolutional neural networks; Computer engineering -- Periodicals; Electrical engineering -- Periodicals; Electrical engineering -- Data processing -- Periodicals; Ordinateurs -- Conception et construction -- Périodiques; Électrotechnique -- Périodiques; Électrotechnique -- Informatique -- Périodiques; Computer engineering; Electrical engineering; Electrical engineering -- Data processing; Periodicals; Electronic journals; 621.302854
• DOI: 10.1016/j.compeleceng.2019.106531
• Languages: English
• ISSNs: 0045-7906

Abstract

Abstract: Recent advances in deep learning have prompted a surge of interest in analysis of medical images. In this study, we developed a convolutional neural network (CNN) for autonomous detection of colorectal polyps, in images captured during wireless colon capsule endoscopy, with risk of malignant evolution to colorectal cancer. Our CNN is an improved version of ZF-Net which uses a combination of transfer learning, pre-processing and data augmentation. We further deployed our CNN as the basis for a Faster R-CNN to localize regions of images containing colorectal polyps. We created an image database of 11, 300 capsule endoscopy images from a screening population, including colorectal polyps (any size or morphology, N=4800) and normal mucosa (N=6500). Our CNN scored an accuracy of 98.0%, a sensitivity of 98.1% and a specificity of 96.3%. Our network outperforms all state-of-the-art results in autonomous detection of colorectal polyps and shows high interpretability in terms of sensitive regions.