Research on Gastritis Endoscopic Images Classification Based on SE-RepLKNet

Autoimmune gastritis (AIG) is an atrophic gastritis caused by abnormal autoimmune function. Recent studies indicate an increasing prevalence of AIG in China. In advanced stages, AIG can progress to neuroendocrine tumors and gastric adenocarcinoma, posing a life-threatening risk to patients. However, due to the insufficient clinical awareness and nonspecific clinical manifestations of the disease, AIG has a low diagnostic rate and is frequently misdiagnosed as type B gastritis caused by Helicobacter pylori infection, which leads to delayed diagnosis. Gastroscopy serves as a critical tool for diagnosing AIG based on distinct gastroscopic features of AIG and type B gastritis. This study proposes a gastritis classification and diagnostic method based on gastroscopic images using a deep learning model, RepLKNet, with large-kernel convolution. Additionally, a channel attention mechanism, the SE module, is integrated to enhance the model's feature extraction capability. Considering the scarcity of publicly available medical image datasets for pretraining, a two-stage pretraining strategy is adopted to incrementally train the model for extracting hierarchical features. Experimental results demonstrate a classification accuracy of 89.6% on gastroscopic images. This system provides clinicians with more accurate classification suggestions for two frequently misdiagnosed types of gastritis, offering crucial information for subsequent treatment and examination.