目的：探究阻塞性冠心病患者中舌象纹理特征与冠周脂肪影像组学的相关性。
方法：收集辽宁中医药大学附属医院131例就诊的疑似冠心病患者，其中阻塞性冠心病63例、非阻塞性冠心病68例，提取临床常规指标、舌象纹理特征及右冠状动脉影像组学参数。独立样本t检验、曼-惠特尼U检验、卡方检验、错误发现率校正用于比较组间差异。使用斯皮尔曼相关性分析舌象纹理特征与冠周脂肪影像组学的相关性。采用弹性网络进行特征筛选,按照7：3的比例随机抽样划分为训练集和测试集，使用BP神经网络模型构建冠周脂肪及冠周脂肪与舌象纹理特征联合模型，使用ROC曲线、准确性、精度、召回率、特异度、F1分数、Kappa系数评估模型。
结果：在疑似冠心病患者中，舌象纹理与冠周脂肪影像组学共检出22组弱负相关、1组中等正相关及23组弱正相关；阻塞性冠心病患者中检出4组中等负相关、4组弱负相关、2组中等正相关及6组弱正相关；非阻塞性冠心病患者中检出1组中等负相关、6组弱负相关、1组中等正相关及10组弱正相关。基于BP神经网络构建诊断模型，联合舌象纹理特征后较单一冠周脂肪影像组学模型测试集AUC有所提升。
结论：舌象纹理特征与冠周脂肪影像组学参数在阻塞性冠心病中存在一定的相关性；舌象纹理信息对冠周脂肪诊断阻塞性冠心病具有增量价值。
Abstract:Objective: To investigate the correlation between tongue image texture features and pericoronary adipose tissue (PCAT) radiomics in patients with obstructive coronary heart disease.
Methods: A total of 131 patients with suspected coronary heart disease admitted to the Affiliated Hospital of Liaoning University of Traditional Chinese Medicine were enrolled, including 63 cases of obstructive coronary heart disease and 68 cases of non-obstructive coronary heart disease. Clinical routine indicators, tongue image texture features, and right coronary artery radiomics parameters were extracted. Independent sample t-test, Mann-Whitney U test, Chi-square test, and false discovery rate correction were used to compare intergroup differences. Spearman correlation analysis was employed to examine the correlation between tongue image texture features and PCAT radiomics. Elastic net was applied for feature selection. The dataset was randomly split into training and test sets at a 7:3 ratio. BP neural network models were constructed using PCAT features alone and in combination with tongue image texture features. Model performance was evaluated using receiver operating characteristic (ROC) curves, accuracy, precision, recall, specificity, F1 score, and Kappa coefficient.
Results: In patients with suspected coronary heart disease, 22 weak negative correlations, 1 moderate positive correlation, and 23 weak positive correlations were identified between tongue image texture features and PCAT radiomics. In patients with obstructive coronary heart disease, 4 moderate negative correlations, 4 weak negative correlations, 2 moderate positive correlations, and 6 weak positive correlations were detected. In patients with non-obstructive coronary heart disease, 1 moderate negative correlation, 6 weak negative correlations, 1 moderate positive correlation, and 10 weak positive correlations were observed. The BP neural network diagnostic model combining tongue image texture features with PCAT radiomics demonstrated improved performance on the test set compared with the PCAT radiomics model alone.
Conclusion: Tongue image texture features exhibit certain correlations with PCAT radiomics parameters in obstructive coronary heart disease, and tongue texture information provides incremental value for the diagnosis of obstructive coronary heart disease when combined with PCAT radiomics.
