Identification of small intestinal bleeding by small intestinal capsule endoscopy with intelligent assistant system based on deep convolutional neural network / 中华消化杂志

ABSTRACT

Objective:To verify the capability of small intestinal capsule endoscopy with intelligent assistant system based on deep convolutional neural network (DCNN) in the identification and diagnosis of small intestinal bleeding.Methods:A total of 158 235 small intestinal capsule endoscopy images of 1 970 patients were collected from ESView platform (including 3 765 images of 165 patients with small intestinal bleeding) for training of DCNN-based small intestinal capsule endoscopy with intelligent assistant system. In the validation phase, the capability of the system in identification and diagnosis of small intestinal bleeding was verified by images of 100 patients with small intestinal bleeding (10 cases of active bleeding, 31 cases of blood clot and 59 cases of submucosal hemorrhage).Results:Small intestinal bleeding lesions could be identified by the DCNN-based intelligent assistant system, and also could be displayed with mark on the original capsule endoscopy images. This system also could mark multiple bleeding images of the same lesion and multiple different bleeding lesions in the same image. With this system the average reading time of 100 cases of small intestinal bleeding of images the doctors used was (5.23±1.31) minutes per case. The sensitivity of the diagnosis of small intestinal bleeding was 99.00% (95% confidence interval 93.76% to 99.95%).Conclusions:The sensitivity of small intestinal bleeding identification by small intestinal capsule endoscopy with DCNN-based intelligent assistant system is high, which can be used to assist image reading doctors to identify and diagnose of small intestinal bleeding.