治疗前多参数MRI影像组学特征预测晚期宫颈鳞癌患者新辅助化疗后淋巴结转移

Radiomics analysis for prediction of lymph node metastasis after neoadjuvant chemotherapy based on pretreatment MRI in locally advanced cervical squamous cell carcinoma

目的基于治疗前多参数磁共振成像(multi-parametric magnetic resonance imaging,mpMRI)影像组学特征,结合临床变量构建模型,预测局部晚期宫颈鳞癌(locally advanced cervical squamous cell carcinoma,LACSCC)患者新辅助化疗(neoadjuvant chemotherapy,NACT)后淋巴结转移状况。材料与方法回顾性分析两个中心2013年1月至2022年2月的265例接受NACT并行根治性子宫切除术的LACSCC患者病例及影像,中心1的数据用于模型训练,中心2的数据用于模型验证。所有患者NACT治疗前行盆腔MRI检查。于矢状位T2加权成像(sagittal T2-weighted imaging,Sag_T2WI)、轴位弥散加权成像(axial diffusion-weighted imaging,Ax_DWI)和延迟期矢状位对比增强T1加权成像(sagittal T1-weighted contrast-enhanced imaging,Sag_T1C)勾画肿瘤感兴趣区(region of interest,ROI)并提取影像组学特征。通过K最佳(K-Best)及最小绝对值收缩与选择算法(least absolute shrinkage and selection operator,LASSO)降维并筛选出与淋巴结转移强相关影像组学特征。基于每个序列筛选后的组学特征构建三个单序列模型,在所有特征间作相关性分析,排除高度相关的组学特征,并对临床变量进行多变量回归分析,融合临床变量及筛选后的影像组学特征构建临床-影像组学的组合模型,比较模型间性能差异。利用受试者工作特征(receiver operating characteristic,ROC)曲线及决策曲线(decision curve analysis,DCA)评估模型的诊断性能及临床效能。结果Sag_T2WI、Ax_DWI、Sag_T1C三个序列分别筛选出6、3、7个与淋巴结转移高度相关的组学特征,其中2个形状特征和10个纹理特征被纳入组合模型。多因素logistic回归分析显示MRI评估的淋巴结状态是淋巴结转移的预测因素(P<0.05)。与单序列模型相比,组合模型具有更好的预测能力,在训练集和验证集的诊断能力最高,ROC曲线下面积、敏感度和特异度分别为0.848[95%(confidence interval,CI):0.785~0.912]、78.2%、74.4%和0.827(95%CI:0.737~0.917)、80.8%、69.4%。DCA显示如果风险阈值超过60%,则用组合模型预测LACSCC患者NACT后淋巴结状态可获得较大的临床效益。结论基于治疗前MRI,联合Sag_T2WI、Ax_DWI、Sag_T1C三个序列的组学特征及临床信息对LACSCC患者NACT后淋巴结转移具有较好的预测效能。

Objective:To establish a radiomics model based on pre-treatment multi-parametric magnetic resonance imaging(MRI)combined with clinical factors for early prediction of lymph node metastasis in patients with locally advanced cervical squamous cell carcinoma(LACSCC)after neoadjuvant chemotherapy(NACT).Materials and Methods:The baseline radiological image and case data of 265 LACSCC patients who received NACT and radical hysterectomy from January 2013 to Febrary 2022 in two centers were retrospectively analyzed.The data of center 1 were used for training,and the data of center 2 were used for validation.All patients underwent pelvic MRI before NACT.Radiomics features were extracted from sagittal T2-weighted imaging(Sag_T2WI),axial diffusion-weighted imaging(Ax_DWI)and sagittal delayed T1-weighted contrast-enhanced imaging(Sag_T1C).The K-Best and least absolute shrinkage and selection operator(LASSO)were used to reduce the dimension and screen out the radiomics features strongly related to lymph node metastasis.Three single-sequence models were constructed based on the radiomics features selected from each sequence.Correlation analysis was performed among all features,excluding highly correlated radiomics features,and multivariate regression analysis was performed on clinical variables,which were combined to construct the clinical-radiomics model.Model performance was compared using receiver operating characteristic(ROC)curves and decision curve analysis(DCA)to evaluate diagnostic performance and clinical efficacy.Results:Six,three,and seven radiomics features were screened out from Sag_T2WI,Ax_DWI,and Sag_T1C sequences,respectively,which were highly correlated with lymph node metastasis,including 4 shape features and 12 texture features,of which 2 shape features and 10 texture features were included in the combined model.Multivariate logistic regression analysis showed that radiological lymph node status(LNr)was a correlative factor of lymph node metastasis(P<0.05).Compared with the single-sequence model,the combined model had better predictive ability and the highest diagnostic ability in the training and validation sets,the area under the curve(AUC)of ROC,sensitivity and specificity were 0.848[95%(confidence interval,CI):0.785-0.912],78.2%,74.4%and 0.827(95%CI:0.737-0.917),80.8%,69.4%,respectively.DCA showed that if the risk threshold exceeded 60%,the combination model could obtain greater clinical benefit in predicting lymph node status of LACSCC patients after NACT.Conclusions:Based on pre-treatment MRI,the combination of the radiomics features of Sag_T2WI,Ax_DWI,and Sag_T1C sequences and clinical information can predict lymph node metastasis after NACT in LACSCC patients.