"智""镜"联合——探索慢性胃炎的中西医诊断与辨证OA

Objective To explore the classification diagnosis and syndrome differentiation of chronic gastritis(CG)based on various machine learning models using gastroscopic images,and to assess its value for theoretical innovation in the diagnosis of Chinese and Western medicine.Methods From June 2023 to March 2024,2,052 gastroscopic images from patients diagnosed with CG were retro-spectively collected at Hebei Provincial Hospital of Traditional Chinese Medicine.After data preprocessing,three machine learning models were established to evaluate the value of machine learning-assisted gastroscopy for diagnosing CG.As an supplement of tradi-tional Chinese medicine(TCM)inspection,gastroscopy was utilized.From the original dataset,2016 gastroscopic images were selected to form a new dataset.Based on machine learning,deficiency-excess syndrome differentiation was performed to explore the role of gas-troscopy in TCM syndrome differentiation of CG.Results The accuracy rates of the three gastroscopy-based models(ViT,ResNet50,VGG16)we trained for chronic atrophic gastritis(CAG)were 98.60%,95.16%,and 95.81%,respectively.For chronic non-atrophic gastritis(CNAG),the accuracies were 99.16%,99.58%,and 99.16%,respectively.In further experiments,the four machine learn-ing models(ViT,ResNet50,VGG16,MobileNet)achieved accuracies of 95.09%,88.24%,90.20%,and 87.25%for identifying the CG excess syndrome group,and 89.10%,88.11%,89.14%,and 91.09%for the CG deficiency syndrome group,respectively.Con-clusion This study demonstrates that machine learning can serve as a reliable method to assist in the gastroscopic diagnosis of CG.Fur-thermore,the gastroscopy-based machine learning diagnostic model we developed can be used for TCM syndrome differentiation of CG.Machine learning can assist physicians in disease classification and syndrome differentiation,offering clinical value for human-computer collaborative decision-making.