CT纹理分析在胰腺导管腺癌术前预测肠系膜上门静脉侵犯中的应用：与CT影像学特征的比较。 CT texture analysis for the presurgical prediction of superior mesenteric-portal vein invasion in pancreatic ductal adenocarcinoma: comparison with CT imaging features
目的：探讨计算机断层扫描（CT）纹理分析（TA）和成像特征在评估胰腺导管腺癌（PDAC）患者疑似手术肠系膜上门静脉（SMPV）侵犯中的价值。
材料和方法：回顾性分析54例胰头或钩突PDAC患者，怀疑SMPV受累。通过手术探查确定SMPV的侵袭状态。对于每个患者，在预处理CT上提取396个纹理特征。使用非参数检验和最小冗余最大相关性进行特征选择。使用多元逻辑回归构建CTTA模型，并计算模型的受试者操作特征下面积（AUROC）。两名评审员独立评估了SMPV侵袭的定性成像特征，并评估了观察者之间的一致性。使用McNemar检验比较了SMPV侵袭的成像特征和CTTA模型的诊断性能。
结果：54例PDAC患者中，23例（42.6%）检测到SMPV侵袭。CTTA模型的AUROC为0.88（95%置信区间，0.76-0.97），特异性（0.90）显著高于两位评审者（0.61和0.65；p=0.027和0.043）。两位评审人之间的观察者间一致性中等（κ=0.517），CTTA模型准确预测11例。
结论：CTTA可以预测PDAC中疑似SMPV的侵袭，可能是定性成像评估的有益补充。
Figure 2. Overview of CT-texture analysis in this study. (a) ROI segmentation. (b) Feature extraction. (c) Feature selection and CTTA model building. The final CTTA model (Rad-score) was constructed by using multivariable logistic regression. The Rad-score was calculated by a linear combination of the significant features (x1 and x2) that were weighted by their respective β coefficients (β1 and β2) obtained from the multivariable analysis. The diagnostic ability of the CTTA model was reported by using the AUROC, accuracy, sensitivity, and specificity.
Figure 3. Radiomics scores and performance comparisons between the CTTA model and the reviewers. (a) Radiomics score (subtraction of the cut-off determined by the Youden index) of each patient. (b) ROC curve of the CTTA model, with the two experienced reviewers' performances for comparison. The CTTA model yielded an AUROC of 0.88 and performed better than both reviewers. (c) Confusion matrices of the CTTA model and the two reviewers, comparing the true labels and the diagnostic labels for invasion and non-invasion.
Figure 4. Typical cases with teardrop sign and mimics (T: tumour). (a) A 70-year-old male patient with PDAC. The SMPV was embedded in the mass in the head of the pancreas and visualised as a teardrop shape (arrow), this case was evaluated as the presence of SMPV invasion by the reviewers. The CTTA model also suggested the presence of SMPV invasion. The tumour and the invaded vessels could not be separated during surgery. (b) A 54-year-old female patient with PDAC. The SMPV is in contact with a mass in the pancreatic head. Disagreement interpreting on the teardrop sign (arrow) caused the discordant evaluation of SMPV invasion between the reviewers. The CTTA model indicated the absence of SMPV invasion. The surgical result identified that the absence of SMPV invasion of this case.
Figure 5. Typical cases misclassified by both the reviewers (T: tumour). (a) A 67-year-old male patient with pancreatic ductal adenocarcinoma (PDAC). The uncinate process of pancreatic head tumour surrounds the SMPV. Both reviewers considered that vascular stenosis (arrow) was present and classified this case as the presence of SMPV invasion; the CTTA model predicted the absence of SMPV invasion. Surgical exploration revealed the tumour was completely separate from the SMPV. (b) A 53-year-old male PDAC patient. The contact angle between the SMPV and the pancreatic head mass is < 180°. Both reviewers considered the blood vessels were not conspicuously stenotic or deformed (arrow) and classified this case as the absence of SMPV invasion; the CTTA model indicated the presence of SMPV invasion. Intraoperative exploration revealed the presence of SMPV invasion.
Chen F, Zhou Y, Qi X, Xia W, Zhang R, Zhang J, Gao X, Zhang L. CT texture analysis for the presurgical prediction of superior mesenteric-portal vein invasion in pancreatic ductal adenocarcinoma: comparison with CT imaging features. Clin Radiol. 2021 May;76(5):358-366. doi: 10.1016/j.crad.2021.01.003. Epub 2021 Feb 10. PMID: 33581837.
|手机版|小黑屋|爱科学
( 粤ICP备19015697号 )
