4D-CT影像组学模型鉴别甲状旁腺良性病变与淋巴结的应用初探

Application of radiomics model based on 4D-CT for preoperative discrimination benign lesions of parathyroid from lymph nodes

目的:探讨基于4D-CT扫描的建立影像组学模型对甲状旁腺腺瘤或增生与颈部淋巴结的鉴别诊断价值。方法:回顾性将80例行术前4D-CT平扫和多期增强扫描且经手术病理证实的甲状旁腺腺瘤或增生患者纳入本研究。在甲状旁腺结节及同侧颈部II区淋巴结内逐层手动勾画感兴趣区,获得其容积感兴趣区(VOI),基于平扫期、动脉期、静脉期和延迟期图像,各提取1262个影像组学特征。在每个单独期相或其组合的基础上,特征选择基于L1-based方法,采用L1正则化的Logistic回归分析分别建立相应的影像组学模型,并进行10折交叉验证。采用ROC曲线下面积(AUC)、敏感性和特异性评价各模型的鉴别诊断效能。结果:分别基于平扫期、动脉期、静脉期、延迟期、平扫+动脉期、平扫+动脉+静脉期及平扫+动脉+静脉+延迟期图像建立7个影像组学模型,各模型鉴别甲状旁腺腺瘤或增生与颈部淋巴结的诊断效能均较高(训练组AUC为0.949~0.962,验证组为0.869~0.904)。在验证集中,基于静脉期的影像组学模型的AUC最大(0.904),相应的鉴别诊断敏感度为0.886、特异度为0.785、符合率为0.835。结论:基于4D-CT平扫和多期增强扫描建立的影像组学模型对鉴别甲状旁腺腺瘤或增生与颈部淋巴结的准确性较高,其中以基于静脉期图像的影像组学模型的诊断效能最佳。

Objective:The purpose of this study was to discriminate parathyroid adenomas/hyperplasia from lymph nodes using radiomics models based on four-dimensional computed tomography(4D-CT).Methods:Preoperative 4D-CT images of eighty patients with parathyroid adenomas/hyperplasia were retrospectively included in this study.1262 radiomic features from each single phase,as well as their combinations,were extracted.Based on each single phase and their combinations,the L1-based method was performed for feature selection,radiomics models were developed by using Logistic Regression with L1 regularization and evaluated by performing the repeated 10-fold cross validation.The performance of the radiomics models was assessed using the area under the receiver-operating characteristic curve(AUC).Results:Seven radiomics models were built based on plain,arterial,venous and delayed phases individually,and on combinations of plain+arterial phases,plain+arterial+venous phases,and plain+arterial+venous+delayed phases,respectively.All of the models showed a relatively high AUC for discriminating parathyroid adenomas/hyperplasia from lymph nodes,ranging from 0.949 to 0.962 in the training cohort and from 0.869 to 0.904 in the validation cohort.In the validation cohort,the model built on a single venous phase showed the highest AUC of 0.904,with sensitivity of 0.886,specificity of 0.785 and accuracy of 0.835.Conclusion:Radiomics models based on 4D CT images may be a helpful approach for discriminating parathyroid adenomas/hyperplasia from lymph nodes.The radiomics model based on a single venous phase showed the best diagnostic efficacy.