基于T2WI联合ADC图的影像组学模型术前预测宫颈癌宫旁组织浸润状况的研究

Research on Preoperative Prediction of Parametrial Tissue Invasion of Cervical Cancer Based on T2WI Combined ADC Imaging

目的探索基于磁共振成像T2WI联合ADC图的影像组学模型对宫颈癌术前宫旁组织浸润的预测价值。方法回顾性收集行根治性子宫切除的137例宫颈癌(2009版FIGO分期IB1~IIA)患者的术前多参数MRI影像和术后病理资料,将这些病例随机分配为训练集(96例)和测试集(41例),应用3D Slicer软件对横轴位肿瘤最大层面图像(T2WI和ADC图)进行人工分割勾画出ROI,在ROI中共提取2096个特征,用Lasso算法筛选出15个有效特征,分别建立训练集和测试集的影像组学特征标签,联合相关临床病理指标建立多变量逻辑回归模型,对模型性能进行测试和验证,并开发相应的Nomogram图。结果影像特征标签分类性能在训练集中AUC为0.935(95%CI 0.887~0.984),测试集中AUC为0.810(95%CI 0.675~0.946)。多变量逻辑回归模型在训练集中C-Index为0.926(95%CI 0.889~0.982),测试集中C-Index为0.875(95%CI 0.826~0.913),H-L拟合优度检验显示预测模型的校正曲线和理想曲线拟合度良好。结论基于磁共振成像T2WI联合ADC图的影像组学模型对术前宫颈癌患者的宫旁组织浸润状况具有良好的预测效能,可以辅助患者术前的个体化治疗决策。

Objective To explore the predictive value of imaging omic model based on MRI T2WI combined with ADC for preoperative prediction of parametrial invasion status of cervical cancer.Methods The preoperative multi-parameter MRI and postoperative pathological data of 137 patients with cervical cancer(FIGO stage IB1-IIA,2009 edition)who underwent radical hysterectomy were retrospectively collected.These cases were randomly assigned to the training cohort(96 cases)and the validation cohort(41 cases).The 3D Slicer software was used to segment the maximum horizontal tumor images(T2WI and ADC maps)manually and delineate the ROI.A total of 2096 features were extracted from the ROIs of all cases.A total of 15 valid features were screened out using Lasso algorithm and the radiomics signatures of the training cohort and the validation cohort were established respectively.Finally,multivariate logistic regression model was established by combining those signatures with related clinicopathological indicators,tested and validate the model performance and developed the corresponding Nomogram.Results The AUC of classification performance of radiomics signatures was 0.935(95%CI,0.887~0.984)in the training cohort and 0.810(95%CI,0.675~0.946)in the validation cohort.The C-index of the multivariate logistic regression model was 0.926(95%CI,0.889~0.982)in the training cohort,and 0.875(95%CI,0.826~0.913)in the validation cohort.The Hosmer-Lemeshow test showed that the correction curve of the prediction model fits well with the ideal curve.Conclusion The preoperative prediction model based on the T2WI combined with ADC maps has a good efficacy in predicting the parametrial invasion status of the patients with cervical cancer,which can assist doctors to make preoperative individualized treatment decisions for patients.