增强CT联合纹理分析对于胰腺导管内乳头状黏液性肿瘤恶性潜能的预测价值

Contrast enhanced CT combined with texture analysis for predicting the malignant potential of intraductal papillary mucinous neoplasm of the pancreas

目的:探讨增强CT图像的纹理分析对胰腺导管内乳头状黏液性肿瘤(Intraductal papillary mucinous neoplasm,IPMN)恶性潜能的预测价值。方法:回顾性收集2010年1月—2019年12月经手术病理证实为分支胰管型IPMN(Branch duct IPMN,BD-IPMN)或混合型IPMN(Mixed type IPMN,MT-IPMN)、术前1月内均行CT增强检查患者103例,分为低风险组67例,高风险组36例。对病灶进行常规影像评估,对两组间有统计学差异的临床及常规影像特征进行多因素Logistic回归分析,确定独立危险因素。使用Frontier后处理软件提取纹理特征,删除相关系数大于0.9的冗余特征,并利用Lasso-Logistic回归进行纹理特征选择。使用筛选的临床及常规影像特征和纹理特征建立Logistic回归预测模型,采用十折交叉验证的方法评估预测模型的泛化能力,并绘制ROC曲线。结果:CA19-9升高、壁结节≥5 mm、胰管截断伴远端胰腺萎缩、淋巴结肿大是预测IPMN恶性潜能的独立危险因素,从1691个纹理特征中筛选出8个系数绝对值大于0.1的特征。临床及常规影像特征建立的预测模型曲线下面积(AUC)为0.777(0.664~0.890),纹理特征建立的预测模型AUC为0.895(0.833~0.957),两者联合建立的预测模型AUC为0.881(0.813~0.948)。结论:基于增强CT图像的纹理分析可以较好地预测IPMN的恶性潜能。

Objective:To explore the predictive value of texture analysis based on contrast enhanced CT images for the malignant potential of intraductal papillary mucinous neoplasm(IPMN)of the pancreas.Methods:A total of 103 patients pathologically confirmed branch duct IPMN(BD-IPMN)or mixed type IPMN(MT-IPMN)who underwent contrast enhanced CT within 1 month before the operation were collected retrospectively from January 2010 to December 2019.The patients were divided into low-risk group of 67 cases and high-risk group of 36 cases.The lesions were evaluated by conventional way of imaging,and multivariable Logistic regression analysis was performed on clinical and conventional imaging features which showed significant differences between the two groups to determine independent predictors.Frontier,post-processing software,was used to extract texture features.Redundant features with correlation coefficient greater than 0.9 were removed and then Lasso-Logistic regression was performed to select texture features.The selected clinical and conventional imaging features and texture features were used to establish prediction models with Logistic regression.10-fold cross validation method was used to evaluate the generalization ability of the prediction models with the ROC curves.Results:Elevated level of CA 19-9,mural nodule≥5 mm,truncation of pancreatic duct with distal pancreatic atrophy,and lymphadenopathy were independent predictors of the malignant IPMN.Eight features with coefficients greater than 0.1 were selected from the 1691 texture features.The prediction model established by clinical and conventional imaging features showed AUC of 0.777(0.664~0.890),the prediction model by texture features showed AUC of 0.895(0.833~0.957),and the combined prediction model showed AUC of 0.881(0.813~0.948).Conclusion:Texture analysis based on enhanced CT images showed better performance for predicting the malignant potential of IPMN.