CT影像组学在腹部淋巴瘤、非转移及转移性淋巴结鉴别中的价值

Value of CT radiomics in differentiating normal,lymphoma and metastatic abdominal lymph nodes

目的:探讨CT影像组学在鉴别腹部淋巴瘤、非转移及转移性淋巴结中的价值。方法:收集95例经病理或临床证实的腹部淋巴瘤、非转移及转移性淋巴结患者(共242枚腹部淋巴结)的临床及影像学资料。95例中淋巴瘤32例,共94枚淋巴结(淋巴瘤组);非转移淋巴结36例,共78枚淋巴结(非转移组);转移性淋巴结27例,共70枚淋巴结(转移组)。95例均行腹部CT增强扫描。在层厚1 mm图像上手动勾画淋巴结,经Radcloud平台提取病灶的影像组学特征,建立淋巴瘤组与转移组、淋巴瘤组与非转移组、转移组与非转移组3组模型,使用3种机器学习算法进行独立的训练和验证,并计算准确率、敏感度、特异度和AUC。结果:淋巴瘤组与转移组,K最近邻(KNN)分类器鉴别效能最好:训练集AUC为0.98,敏感度0.95,特异度0.88;验证集AUC为0.98,敏感度0.93,特异度0.95。淋巴瘤组与非转移组,支持向量机(SVM)分类器鉴别效能最好:训练集AUC为0.99,敏感度0.98,特异度1.00;验证集AUC为0.96,敏感度0.97,特异度0.88。转移组与非转移组,线性逻辑回归(LR)分类器鉴别效能最好:训练集AUC为0.96,敏感度0.92,特异度0.87;验证集AUC为0.76,敏感度0.62,特异度0.71。结论:基于CT增强扫描的影像组学特征在鉴别腹部淋巴瘤、非转移及转移性淋巴结方面显示出良好的价值,可为治疗提供更准确信息。

Objective:To explore the value of CT radiomics in distinguishing three kinds of abdominal lymph nodes,including normal lymph nodes,lymphoma and metastatic lymph nodes.Methods:A total of 95 patients with pathologically or clinically confirmed 242 abdominal lymph nodes who underwent abdominal enhanced CT scanning were analyzed retrospectively.They were A total of 242 abdominal lymph nodes were divided into three groups,32 lymphoma patients with 94 lymph nodes(the lymphoma group),36 abdominal tumor patients with 78 non-metastasis enlarged lymph nodes(the non-metastasis group)and 27 patients with 70 metastasis lymph nodes(the metastasis group).The ROI was manually sketched,the imaging features of the lesions were extracted through the Radcloud platform,three machine learning algorithms were used for independent training and testing,and the accuracy,sensitivity,specificity,and AUC were determined.Results:For the lymphoma group and the metastasis group,the prediction efficiency of the K-nearest neighbor(KNN)classifier was the best,with the AUC,sensitivity,specificity for the training group of 0.98,0.95,0.88,and for the validation group of 0.98,0.93,0.95,respectively.For the lymphoma group and the non-metastasis group,the prediction efficiency of the support vector machine(SVM)classifier was the best,with the AUC,sensitivity,specificity for the training group of 0.99,0.98,1.00,and for the validation group of 0.96,0.97,0.88,respectively.For the metastasis group and the non-metastasis group,the prediction efficiency of the logistic regression(LR)classifier was the best,with the AUC,sensitivity,specificity for the training group of 0.96,0.92,0.87,and for the validation group of 0.76,0.62,0.71,respectively.Conclusions:The imaging characteristics based on CT enhanced scanning are feasible in predicting benign and malignant abdominal lymph nodes and classification,which is expected to provide new technical support for clinical diagnosis and treatment.