CT纹理分析对胰腺导管腺癌分化程度预测效能的研究

Pancreatic Ductal Adenocarcinoma:A CT Texture Analysis Study of Prediction Efficacyin Histopathological Differentiation

目的探索基于术前增强CT的纹理分析对胰腺导管腺癌(PDAC)分化程度的预测效能。方法搜集经病理确诊为PDAC的患者242例,包括中-高分化60例和低分化182例。分别记录患者的性别、年龄、肿瘤部位、肿瘤最大径、CA19-9值、基于常规影像评估的T分期、动脉侵犯及静脉侵犯情况等临床及常规影像特征,进行Logistic回归分析。将242例患者随机分为训练集(n=182)和测试集(n=60),分别在术前增强CT的动脉期和静脉期图像上对肿块进行手动全层勾画,并提取纹理特征。在训练集中进行特征降维,分别构建包括动脉期纹理、静脉期纹理、动脉期及静脉期纹理的三个放射组学分数的计算公式。绘制训练集及测试集受试者工作特征曲线(ROC)评估各放射组学分数的预测效能。结果临床及常规影像特征对PDAC分化程度无预测意义。单独动脉期及单独静脉期各筛选出3个特征,联合动脉期及静脉期共筛选出16个特征,包括5个动脉期特征和11个静脉期特征。联合动脉期及静脉期构建的放射组学分数的训练集及测试集的ROC曲线,其曲线下面积(AUC)分别为0.80、0.79,准确率分别为0.739、0.783,均高于单独动脉期(AUC分别为0.70、0.68,准确率分别为0.678、0.667)或单独静脉期(AUC分别为0.64、0.68,准确率分别为0.656、0.667)。结论PDAC术前增强CT的纹理分析能够预测PDAC的分化程度,基于动脉期与静脉期图像联合的纹理分析预测效能较单一期相好。

Objective To investigate predictive efficacy of the texture analysis based on preoperative enhanced CT in the histopathological differentiation of pancreatic ductal adenocarcinoma.Methods A total of 242 cases of pancreatic ductal adenocarcinoma(PDAC)diagnosed by pathology were enrolled in this retrospective study.60 cases were moderately-well differentiated and 182 cases were low differentiated.Clinical and routine imaging features including gender,age,location of the tumor,maximum diameter of the tumor,CA19-9,T stage assessed based on conventional images,arterial invasion and venous invasion were collected for logistic regression analysis.All 242 cases were randomly divided into training set(n=182)and test set(n=60).Region of interest(ROI)was delineated on the whole area of the mass based on both arterial phase and venous phase of the preoperative enhanced CT,then texture features were extracted.The features were selected in the training set.Then three radiomic scores including arterial phase textures,venous phase textures,arterial phase texture and venous phase textures were built.The predictive efficiency of the radiomic scores wereevaluated by drawing the receiver operating characteristic curve(ROC)of the training set and test set.Results The clinical and routine imaging features had no predictive significance for PDAC histopathological differentiation.3 texture features were selected in arterial phase and 3 texture features were selected in venous phase.With combination of arterial phase and venous phase,a total of 16 texture features were selected,including 5 features from arterial phase and 11 features from venous phase.The ROC of radiomic scores of the training set and the test set constructed in combination of the arterial and venous phases showed an AUC of 0.80,0.79,and an accuracy of 0.74 and 0.78,whichwere higher than arterial phase(AUC were 0.70,0.68,accuracy were 0.68,0.67)or venous phase(AUC were 0.64,0.68,accuracy 0.66,0.67).Conclusion Texture analysis based on the preoperative enhanced CT can predict the histopathological differentiation of pancreatic ductal adenocarcinoma.Texture analysis combining arterial phase and venous phase has better predictive efficacy than that of single phase.