Real-time semantic segmentation of gastric intestinal metaplasia using a deep learning approach

• DOI: 10.5946/ce.2022.005
• Publication Date: 2022-05-09T02:24:10Z
• Authors: Vitchaya Siripoppohn; Rapat Pittayanon; Kasenee Tiankanon; Natee Faknak; Anapat Sanpavat; Naruemon Klaikaew; Peerapon Vateekul; Rungsun Rerknimitr
• Keywords: Intestinal metaplasia

Previous artificial intelligence (AI) models attempting to segment gastric intestinal metaplasia (GIM) areas have failed be deployed in real-time endoscopy due their slow inference speeds. Here, we propose a new GIM segmentation AI model with speeds faster than 25 frames per second that maintains high level of accuracy.Investigators from Chulalongkorn University obtained 802 histological-proven images for training. Four strategies were proposed improve the accuracy. First, transfer learning was employed public colon datasets. Second, an image preprocessing technique contrast-limited adaptive histogram equalization produce clearer areas. Third, data augmentation applied more robust model. Lastly, bilateral network real time. The results analyzed using different validity values.From internal test, our achieved speed 31.53 second. detection showed sensitivity, specificity, positive predictive, negative accuracy, and mean intersection over union values 93%, 80%, 82%, 92%, 87%, 57%, respectively.The combined learning, equalization, can provide sensitivity good accuracy segmentation.