预测浸润性乳腺癌无病生存期的临床影像融合模型研究

Clinical radiomics fusion model for predicting disease-free survival of invasive breast cancer

目的构建术前临床影像组学融合模型,探讨其预测浸润性乳腺癌无病生存期的效能。方法收集2015年7月至2017年5月浙江中医药大学附属第一医院术前接受乳腺动态增强MRI检查的浸润性乳腺癌女性患者131例。按照7∶3的比例随机分为训练集91例与内部验证集40例。从肿瘤和瘤体向外扩5mm区域分别提取原始肿块和瘤内加瘤周的影像组学特征。运用F检验及logistic回归分析对影像组学特征进行缩减后,分别建立影像组学模型。采用多因素Cox比例风险模型分析临床病理风险因素与无病生存期的关系,并确定有效的预测因素建立临床模型。将最优影像组学模型的影像组学评分(Radscore)和独立临床病理风险因素结合建立融合模型。采用ROC曲线评估影像组学模型和融合模型的预测效能。结果在验证集中,临床模型、原始肿块影像组学模型、瘤内加瘤周影像组学模型及融合模型的AUC(95%CI)分别为0.791(0.690~0.893)、0.898(0.839~0.956)、0.930(0.878~0.983)、0.969(0.941~0.996),灵敏度分别为0.789、0.895、0.842、0.947,特异度分别为0.714、0.821、0.857、0.938,其中融合模型预测效能最佳。结论基于术前MRI影像组学特征的临床影像融合模型对浸润性乳腺癌患者的复发风险具有较高的预测效能,对乳腺癌无病生存期的预测和个性化治疗具有一定的价值。

Objective To construct a preoperative clinical radiomics fusion model and evaluate its efficacy in predicting disease-free survival of invasive breast cancer.Methods A total of 131 female patients with invasive breast cancer who underwent dynamic enhancement MRI before surgery in The First Affiliated Hospital of Zhejiang Chinese Medical University from July 2015 to May 2017 were retrospectively analyzed.Patients were randomly divided into a training set(n=91)and a validation set(n=40)according to the ratio of 7:3.The imaging features of the original mass and the intratumoral plus peritumoral images were extracted from the tumor and the tumor expanded outward by 5 mm,respectively.After using F test and logistic regression analysis to reduce the radiomics features,the radiomics models were established respectively.Multivariate Cox proportional risk model was used to analyze the relationship between clinicopathological factors and diseasefree survival,and to identify effective predictors to establish a clinical model.The fusion model was established by combining the Radscore of the optimal radiomics model with independent clinicopathological risk factors.ROC curve was used to evaluate the predictive efficiency of radiomics models and fusion model.Results In the validation set,the ROC curve AUC(95%CI)of the clinical feature model,original tumor radiomics model,intratumoral plus peritumoral radiomics model,and fusion model were 0.791(0.690-0.893),0.898(0.839-0.956),0.930(0.878-0.983),and 0.969(0.941-0.996),respectively.The sensitivity was 0.789,0.895,0.842,and 0.947,respectively,and the specificity was 0.714,0.821,0.857,and 0.938,respectively,among which,the fusion model showed the best predictive efficiency.Conclusion The clinical radiomics fusion model based on preoperative MRI imaging features has high predictive efficacy for the recurrence of patients with invasive breast cancer,and has certain value for the prediction of disease-free survival and personalized treatment of breast cancer.