中医药信息2025，Vol.42Issue(8)：34-39,6.DOI:10.19656/j.cnki.1002-2406.20250806
基于舌象图像深度特征与支持向量机融合的肺结节风险分层预测研究
Risk Stratification Prediction of Pulmonary Nodules Based on Fusion of Tongue Image Deep Features and Support Vector Machine
摘要
Abstract
Objective:To implement risk prediction of pulmonary nodules in different age groups using a Support Vector Machine(SVM)model trained with tongue image features based on deep transfer learning.Methods:Data including chest CT scans,clinical records,corresponding tongue images,and pulmonary nodule diagnoses from 319 cases(128 healthy controls[40.1%]and 191 pulmonary nodule patients[59.9%])were collected from the medical imaging database of The First Affiliated Hospital of Zhejiang Chinese Medical University(Zhejiang Provincial Hospital of Chinese Medicine)between January 1,2024 and February 29,2024.Each group was stratified into subgroups aged≥50 years and<50 years.Three deep transfer learning algorithms(AlexNet,DenseNet,ResNet50)were employed to extract deep features from tongue images.LASSO algorithm reduced dimensionality of extracted features for each subgroup,which were then input into SVM classifiers for model training.AUC values served as primary evaluation metrics,supplemented by sensitivity,specificity,and accuracy to assess model performance and clinical significance.Results:All subgroups'deep features were reduced to 10 key features.Age≥50 years showed statistical significance in discriminating pulmonary nodules(P=0.003<0.01).In the<50 subgroup,SVM models trained with ResNet50-extracted features achieved optimal performance(test AUC=0.85,training AUC=0.96).For≥50 subgroup,AlexNet-based models performed best(test AUC=0.728,training AUC=0.924).Conclusion:The AlexNet-feature SVM model demonstrates superior pulmonary nodule risk prediction for≥50 subgroup,while ResNet50-feature SVM model excels in<50 subgroup,potentially supporting future intelligent TCM tongue diagnosis for pulmonary nodule screening.关键词
肺结节/舌象图像/深度学习/支持向量机/特征提取/风险预测Key words
Pulmonary nodules/Tongue image/Deep learning/Support vector machine/Feature extraction/Risk prediction引用本文复制引用
陈宇辰,张昊坤,单婉婷,聂明毅,林江南..基于舌象图像深度特征与支持向量机融合的肺结节风险分层预测研究[J].中医药信息,2025,42(8):34-39,6.基金项目
国家级大学生创新创业训练计划项目(202310344005X) （202310344005X）
