A preliminary study on establishment of a benign and malignant pathological diagnosis model of gastric tissue based on deep learning

FA Liangling; HUANG Zhen; ZHOU Shan; SUN Wei

doi:10.3969/j.issn.1007-8096.2025.07.008

Abstract：Objective The pathological diagnosis model of benign and malignant gastric tissue was studied based on deep learning technology.Methods A total of 360 patients with gastric diseases admitted to our hospital from January 2016 to June 2024 were selected and divided into benign group(189 cases)and malignant group(171 cases)according to pathological examination results.The clinical data of the two groups were compared.The types of lesions were identified.The morphological parameters of glandular and mucosal epithelium in different pathological tissues were compared.The factors influencing gastric malignancy were analyzed by Logistic regression.Multiparameter magnetic resonance imaging(MRI)radiomics features were screened.The performance of different deep learning models were compared in predicting benign and malignant gastric tissue.The predictive efficiency of clinical model,radiomics model,deep learning model and fusion model was evaluated.Results There were significant differences in age,malnutrition,ferritin,alpha-fetoprotein(AFP),carcinoembryonic antigen(CEA),cancer antigen 211(CA211),CA50,CA125,CA199,CA242 and CA724 between the two groups(P<0.05).Age,malnutrition,ferritin,AFP,CEA,CA211,CA50,CA125,CA199,CA242 and CA724 were independent risk factors for gastric cancer.The recognition accuracy,recall rate,precision,and F1 score of the ResNet model were better than those of GoogLeNet and AlexNet.The four kinds of models had high discriminability,good accuracy and effectiveness,and the fusion model had the highest predictive efficiency.Conclusion The fusion model based on clinical features,multiparametric MRI radiomics and deep learning can accurately diagnose benign and malignant gastric tissue pathology and provide personalized prediction results.