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OA北大核心CSTPCD

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.