Image Classification to Diagnose Chronic Atrophic Gastritis Based on Squeeze-and-Excitation Block

Chronic atrophic gastritis (CAG) is an organ-specific autoimmune disease, and its occurrence and development seriously threaten human health. Therefore, CAG is medically characterized as a serious precancerous lesion. We used a convolutional neural network (CNN) to construct a deep learning network, which helps improve the accuracy of diagnosing CAG. In this paper, we used the INPAINT_TELEA algorithm to remove the watermark in gastric antrum images. According to the characteristics of the dataset, a 14 layers’ convolutional neural network (SE-CAG) was designed, in which the BN layer and scale layer are added to prevent overfitting. The designed network is integrated into the Squeeze-and-Excitation (SE) block. Through the feature reuse principle of SE block, the effective features are highlighted, to realize the classification of our datasets. From experimental results on our dataset, the detection rate of chronic atrophic gastritis is 88.15%, which makes it possible to diagnose CAG automatically. Compared with the ResNet (Deep Residual Network) and VGG16, on our own dataset, the proposed model shows a better performance in diagnosing CAG. Then, we used the FP-Growth algorithm to explore the relationship between atrophic gastritis and other gastroscopic symptoms, so that it can establish the relationship between endoscopy and diagnosis results.