基于APTw的影像组学术前预测宫颈癌淋巴血管间隙侵犯

The radiomics model based on APT for preoperative prediction of cervical cancer lymphovascular space invasion

目的探讨酰胺质子转移加权成像(amide proton transfer weighted imaging,APTw)的影像组学术前预测宫颈癌淋巴血管间隙侵犯(lymphovascular space invasion,LVSI)的价值。材料与方法回顾性分析经手术病理证实的宫颈癌患者病例及影像资料66例。所有患者均行盆腔3.0 T MRI检查,包括轴位T2WI、矢状位T2WI、动态对比增强磁共振成像(dynamic contrast enhanced magnetic resonance imaging,DCE-MRI)和3D-APTw序列扫描。在APTw-T2WI融合图像上对肿瘤实质区域进行感兴趣区(region of interest,ROI)勾画并记录APT值。在APT重建图像上进行肿瘤病灶分割并提取影像组学特征。采用组内相关系数(intra-class correlation coefficient,ICC)选取观察者内和观察者间复测信度好的影像组学特征(ICC>0.900)。采用递归特征消除法(recursive feature elimination,RFE)及最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)算法进行特征降维和筛选。基于logistic回归分类器构建临床模型、APTw影像组学模型和联合组学模型。采用受试者工作特征(receiver operating characteristic,ROC)曲线和决策曲线分析(decision curve analysis,DCA)评估模型的诊断效能和临床价值,采用DeLong检验比较不同模型的预测效能。结果在训练集中,APTw影像组学模型预测宫颈癌LVSI的效能高于临床模型(AUC=0.826 vs.0.675),差异有统计学意义(DeLong检验P<0.05)。联合组学模型在训练集和测试集中的AUC值分别为0.838和0.825。DeLong检验结果显示,联合组学模型在训练集中术前评估LVSI的效能显著高于临床模型和APTw影像组学模型(P均<0.05)。决策曲线显示APTw影像组学模型和联合组学模型在训练集和测试集中均具有较高的临床价值。结论基于APTw的影像组学模型在术前预测宫颈癌LVSI方面具有较高的潜力,联合临床因素能进一步提高预测效能,有望为宫颈癌患者的个体化治疗和预后评估提供重要的支持。

Objective:To explore the value of amide proton transfer weighted imaging(APTw)radiomics in the preoperative assessment of lymphovascular space invasion(LVSI)in cervical cancer.Materials and Methods:Retrospective analysis of 66 cases of pathologically confirmed cervical cancer and their imaging data.All patients underwent pelvic 3.0 T MRI examination,including axial T2WI,sagittal T2WI,dynamic contrast-enhanced magnetic resonance imaging(DCE-MRI),and 3D-APTw sequence scanning.Region of interest(ROI)within the tumor parenchyma were delineated on the APTw-T2WI fusion images,and APT values were recorded.Tumor lesions were segmented on the reconstructed APTw images,and radiomics features were extracted.Intra-class correlation coefficient(ICC)was employed to select radiomics features with good test-retest reliability both intra-and inter-observer assessments(ICC>0.900).Recursive feature elimination(RFE)and least absolute shrinkage and selection operator(LASSO)algorithms were employed for feature dimensionality reduction and selection.A clinical model,APTw radiomics model and combined model were constructed based on logistic regression classifier.The diagnostic performance and clinical utility of the models were evaluated using receiver operating characteristic(ROC)curves and decision curve analysis(DCA).The predictive performance of different models was compared using the DeLong test.Results:In the training set,the APTw radiomics model demonstrated higher efficacy in predicting cervical cancer LVSI compared to the clinical model(AUC=0.826 vs.0.675),with statistically significant differences(DeLong test P<0.05).In the training set and the test set,the AUC values of the combined model were 0.838 and 0.825,respectively.DeLong test results showed that the combined model significantly outperformed the clinical model and APTw radiomics model in preoperative assessment of LVSI in the training set(all P<0.05).The decision curve demonstrated that the APTw radiomics model and the combined model exhibit higher clinical utility in both the train and test sets.Conclusions:The radiomics model based on the APTw shows great potential in preoperatively predicting the LVSI status of patients with cervical cancer.Integration with clinical factors further enhances predictive performance,holding prospects to provide crucial support for individualized treatment and prognosis assessment of cervical cancer patients.