基于影像组学模型预测胶质母细胞瘤患者术后一年死亡风险  

Prediction of One⁃Year Postoperative Mortality Risk of Glioblastoma Patients Based on Radiomics Model

作　　者：李守斌 汪军[1] 张书海 代兴亮 侯唯姝[1] LI Shoubin;WANG Jun;ZHANG Shuhai(Department of Radiology,the First Affiliated Hospital of Anhui Medical University,Hefei,Anhui Province 230022,P.R.China)

机构地区：[1]安徽医科大学第一附属医院放射科,合肥230022 [2]安徽医科大学第一附属医院神经外科,合肥230022

出　　处：《临床放射学杂志》2024年第4期523-527,共5页Journal of Clinical Radiology

基　　金：安徽省自然科学基金项目面上项目(编号:2208085MH251)。

摘　　要：目的 开发新的临床影像组学列线图模型,用于对胶质母细胞瘤患者术后1年死亡风险进行预测,为胶质母细胞瘤患者的预后提供指导。方法 纳入100例胶质母细胞瘤患者,分成训练组(n=50)和验证组(n=50)。开发一种纳入患者相关临床因素和MRI影像组学特征的列线图。提取患者增强扫描感兴趣区的影像组学特征,通过最小绝对收敛和选择算子(LASSO)筛选出用于建模的特征,并计算出影像组学评分。单因素多因素分析寻找有意义的临床变量。建立单独基于影像组学评分构建的单一模型以及基于临床变量和影像组学评分构建的联合模型。在训练组和验证组中分别使用ROC曲线、临床决策曲线对两种模型进行评价,筛选出最优模型用于指导临床。结果 从胶质母细胞瘤患者增强图像的感兴趣区筛选出有意义的影像组学特征并计算出影像组学评分。从患者相关临床因素中筛选出2个独立危险因素,分别为病灶水肿程度和病灶有无出血。联合模型的AUC在训练组和验证组中分别为0.890和0.879。单一模型的AUC在训练组和验证组中分别为0.842和0.798。临床决策曲线说明模型具有较高的临床实用性。结论 联合模型优于单一模型。结合影像组学评分和相关临床变量构建的模型准确性较高。临床影像组学的列线图是一种非侵入型预测工具,可以辅助临床医师评估胶质母细胞瘤患者术后1年死亡风险。Objective To develop a new clinical radiomic nomogram model for predicting the one⁃year mortality risk of glioblastoma patients after surgery,to provide guidance for the prognosis of glioblastoma patients.Methods 100 patients with glioblastoma were enrolled and divided into training group(n=50)and verification group(n=50).A graph was de⁃veloped incorporating patient⁃relevant clinical factors and MRI radiomic features.The radiomic features of the enhanced re⁃gion of interest were extracted,and the features for modeling were screened by the minimum absolute convergence and selec⁃tion operator(LASSO regression),and the radiomic score was calculated.Univariate and multivariate analyses were conduc⁃ted to find meaningful clinical variables.A single model based solely on radiomic score and a combined model based on clinical variables and radiomic score were established.ROC curve,clinical decision curve was respectively used to evaluate the two models in the training group and the verification group,and the optimal model was selected for clinical guidance.Results Significant radiomic features were screened from the regions of interest in the enhanced images of glioblastoma pa⁃tients and radiomic scores were calculated.Two independent risk factors were screened from related clinical factors of pa⁃tients,namely the degree of edema in the lesion and the presence or absence of bleeding in the lesion.The AUC of the com⁃bined model was 0.890 in the training group and 0.879 in the verification group.The AUC of the single model was 0.842 in the training group and 0.798 in the verification group.Clinical decision curve showed that the model had high clinical prac⁃ticability.Conclusion The combined model is superior to the single model.The accuracy of the model combined with ra⁃diomic scores and relevant clinical variables was higher.The clinical radiomic nomogram is a non⁃invasive predictive tool to assist clinicians in assessing the one⁃year risk of death in patients with glioblastoma.

关 键 词：磁共振成像 影像组学 列线图 神经胶质瘤 

分 类 号：R445.2[医药卫生—影像医学与核医学] R739.41[医药卫生—诊断学]