Image Auxiliary Diagnosis System for Digestive Tract Diseases Based on Convolutional Neural Network

Authors

  • Liangang Peng, Bin Tang, Song Feng, Jie Peng

Abstract

At present, most researches on intelligent assisted diagnosis of gastrointestinal diseases such as early gastric cancer choose magnified endoscopic images as training samples, and there are relatively few studies using white light gastroscopic images. However, the distribution of medical resources in China is uneven, and white light gastroscopy is still the most effective way for early gastric cancer screening. At the same time, there are insufficient endoscopists and large differences in diagnosis. In view of the above-mentioned status, it is proposed to construct an image intelligent auxiliary diagnosis system based on convolutional neural network. First, the white-light gastroscopy image samples are preprocessed and data enhanced; second, the classification effects of the three convolutional neural network models of AlexNet, VGGNet and ResNet are compared, and the structure of the ResNet network is optimized. The experiment shows that the optimized ResNet network the effect is the best, and the average F1 of the two-class recognition model for early gastric cancer disease reaches 94.87%; then, a prospective verification experiment is designed, and the optimized ResNet network performs the best; finally, in order to give full play to the clinical value of the model, design and develop a digestive endoscopy image auxiliary diagnosis system.

Published

2020-02-28

Issue

Section

Articles