中国中西医结合影像学杂志

目的：探讨原发性肝癌DWI影像组学特征与中医证型的相关性及其在中医辨证分型中的应用价值。方法：回顾性收集原发性肝细胞癌患者70例，根据中医诊断标准分为肝癌虚证和实证，均在治疗前行MRI增强扫描。在DWI图像上手动勾画ROI并提取纹理特征，共包含1 316个纹理特征，采用单因素Logistic回归分析、方差分析和秩和检验方法对1 316个纹理特征进行降维，建立K-近邻算法（KNN）影像组学模型。70例以7∶3的比例随机分配成训练组和验证组进行模型内部验证，采用ROC曲线评估模型的性能。结果：70例中经中医证型判断虚证34例（48.6%），实证36例（51.4%）。经特征降维后最终筛选出4个影像组学特征与肝癌虚、实证有相关性（r值0.368～0.457，均P<0.05）。基于筛选的4个影像组学特征构建KNN影像组学模型，该模型在训练组中的AUC值为0.831、准确率77.1%、敏感度72.0%、特异度82.6%，在验证组中的AUC值为0.773、准确率63.6%、敏感度63.6%、特异度63.6%。结论：原发性肝癌中医证型相关的4个影像组学特征及其构建的影像组学模型，在原发性肝癌虚、实证判断方面具有较大的应用潜能，有助于中医辨证施治的客观化、标准化。

Objective:To explore the correlation between DWI radiomics features and TCM syndromes and the value of the TCM syndrome differentiation typing of primary hepatocellular carcinoma(HCC). Methods:A total of 70 patients with primary HCC were retrospectively collected,including 53 males and 17 females,with an average age of(61.2±10.3) years old. According to the diagnostic criteria of TCM,all the enrolled patients were divided into deficiency syndrome group and excess syndrome group,and examined by enhanced MRI exam. The ROI was manually sketched and the texture features were extracted from DWI image. 1 316 texture features were reduced dimension by univariate logistic regression analysis,variance analysis and rank sum test,and K nearest neighbor(KNN) predictive radiomics model was established. The performance of the model was evaluated by ROC curve. The training group and the test group were randomly assigned as 7 :3 ratio. Results:In 70 cases of primary HCC,34 cases(48.6%) were the deficiency syndrome and 36 cases(51.4%) were the excess syndrome,which confirmed by TCM syndrome. A total of four radiomics features were selected through dimension reduction steps,and all of them were statistically correlated with the deficiency syndrome and the excess syndrome(r of 0.368～0.457,P<0.05). The KNN model was constructed based on four selected radiomics features. The AUC value,accuracy,sensitivity and specificity of the model in the training group were 0.831,77.1%,72.0% and 82.6%,respectively,and in the test group,they were 0.773,63.6%,63.6%,and 63.6%,respectively.Conclusions:The model based on the four radiomics features extracted from DWI images has great potential application in the deficiency and excess syndrome of primary HCC,which is helpful to make TCM syndrome differentiation and treatment more objective and standardized.