Abstract

Info

Author(s):: GONG Jiamin1; MA Doudou1; JIANG Jiewei1; ZHANG Yaqiong2; PEI Mengjie3; 1. School of Electronic Engineering, Xian University of Posts and Telecommunications, Xian 710121, China 2. Department of Gastroenterology, Shanxi Provincial Peoples Hospital of Shanxi Medical University, Taiyuan 030012, China 3. School of Computer Science & Technology, Xian University of Posts and Telecommunications, Xian 710121, China

Keywords:: chronic atrophic gastritis deep learning Squeeze_and_ExcitaionApriori algorithm

Abstract:: Chronic atrophic gastritis is a common stomach disease, and it may develop into gastric cancer if without timely treatment. However, the sensitivity of gastroscopy in the examination of atrophic gastritis is only about 42%, and the biopsy is affected by many factors. Therefore, convolutional neural network is used to improve the diagnosis accuracy of chronic atrophic gastritis. INPAINT_TELEA algorithm is firstly used to preprocess the image of the gastric antrum for removing the watermarks in the image, and then residual network is improved and embed into Squeeze_and_Excitation module to realize the diagnosis of chronic atrophic gastritis. Finally, the improved network (SR-CAGnet) is applied to enhance the classification effect of images by establishing a short circuit mechanism and adopting a feature recalibration strategy. The results show that the detection rate of chronic atrophic gastritis by SR-CAGnet reaches 87.92% as compared with Alexnet and improved ResNet, which indicates the proposed algorithm has a good performance on recognition. Through the analysis by Apriori algorithm, the relationship between atrophic gastritis and other symptoms detected by gastroscopy is obtained, thus assisting doctors diagnosis. Finally, the validity of the model is verified using CAM heat map.