基于增强CT影像组学鉴别胰腺癌与肿块型慢性胰腺炎的价值

Differentiating pancreatic cancer from mass-forming chronic pancreatitis based on enhanced CT radiomics

目的探讨增强CT影像组学对胰腺癌(PC)与肿块型慢性胰腺炎(MFCP)的鉴别意义。方法选取92例PC患者和61例MFCP患者分别作为PC组和MFCP组,收集2组患者的增强CT扫描影像资料,采用3D Slicer 4.8.1软件与AK软件分别分割与提取病灶影像的组学特征,按照7∶3随机分层抽样方法分为106例训练组(42例MFCP+64例PC)和47例验证组(19例MFCP+28例PC),对训练组进行特征筛选及降维,获得MFCP与PC之间存在显著差异的最优特征子集,采用逻辑回归机器学习算法以最优特征建立预测模型,绘制受试者操作特征曲线(ROC)、并计算ROC曲线下面积(AUC),验证预测模型对鉴别MFCP与PC的AUC、准确度、敏感度、特异度。结果患者的CT增强动脉期图像总共提取了1037个影像组学特征,获得鉴别PC与MFCP最具有显著差异的7个特征参数;训练组患者的影像组学特征构建的预测模型对鉴别MFCP与PC的AUC、准确度、敏感度、特异度分别为0.967、0.905、0.889及0.929,验证组患者AUC、准确度、敏感度、特异度分别0.968、0.787、0.679及0.947。结论基于增强CT影像组学模型有助于鉴别PC和MFCP。

Objective To explore the differential significance of enhanced CT radiomics in pancreatic cancer(PC)and mass-forming chronic pancreatitis(MFCP).Methods A total of 92 patients with PC and 61 patients with MFCP were selected as PC group and MFCP group,respectively.The enhanced CT scan data of the two groups were collected,using 3D Slicer 4.8.1 software and AK software to segment and extract the radiomics characteristics of lesion respectively,According to 7∶3 random stratified sampling method,106 cases were divided into training group(42 cases of MFCP+64 cases of PC)and 47 cases were divided into validation group(19 cases of MFCP+28 cases of PC).In the training group,feature selection and dimension reduction were carried out to obtain the optimal feature subset with significant differences between MFCP and PC,logistic regression machine learning algorithm was used to establish the prediction model based on the optimal features,and the differential diagnostic performance of the model was evaluated by the receiver operating characteristic curve(ROC).Finally,independent verification was performed on the validation group data.Results A total of 1037 radiomics features were extracted from the CT enhanced arterial phase images of the patients,and seven feature parameters with the most significant difference between PC and MFCP were obtained.The AUC,accuracy,sensitivity,and specificity of the prediction model constructed from the radiomics features of the training group were 0.967,0.905,0.889,and 0.929,respectively,while the AUC,accuracy,sensitivity and specificity of the validation group were 0.968,0.787,0.679,and 0.947,respectively.Conclusion The radiomics model based on enhanced CT is helpful to distinguish PC and MFCP.