磁共振波谱联合减影技术在高级别胶质瘤影像组学分级预测的研究

MRS combined with subtraction technique in the prediction of high grade glioma radiomics grading

目的 在常规影像组学分析中引入磁共振波谱(magnetic resonance spectroscopy, MRS)和减影技术,鉴别高级别胶质瘤分级。材料与方法 回顾性分析60例经病理证实为高级别脑胶质瘤患者(Ⅲ级25例、Ⅳ级35例)的影像数据,基于对比增强T1加权成像(contrast-enhanced T1-weighted imaging, CE-T1WI)图像以及减影图像提取形状、纹理特征等影像组学特征,使用t检验、带有交叉验证的最小绝对收缩和选择算子进行特征筛选,结合MRS的3种代谢物峰值比,使用随机森林算法构建高级别胶质瘤分级鉴别模型并评估模型性能。结果 基于CE-T1WI图像构建的模型的测试集曲线下面积(area under the curve, AUC)为0.78;基于减影图像构建的模型的测试集AUC为0.81;基于MRS的代谢物峰值比构建的模型的测试集AUC为0.80;基于CE-T1WI图像及MRS的3种代谢物峰值比构建的模型的测试集AUC为0.95。结论 基于CE-T1WI图像、减影图像及MRS序列的影像组学,在鉴别Ⅲ级和Ⅳ级胶质瘤方面均具有较好的表现,在单序列模型中,减影模型表现最好,在联合序列模型中,CE-T1WI联合MRS模型表现最好。多模态影像组学分析可为鉴别Ⅲ级和Ⅳ级胶质瘤提供有效临床辅助。

Objective:Magnetic resonance spectroscopy(MRS)and subtraction techniques were introduced in the conventional imaging omics analysis to identify high-grade glioma grade.Materials and Methods:The imaging data of 60 patients with pathologically confirmed high-grade glioma(25 patients with gradeⅢand 35 patients with gradeⅣ)were retrospectively collected,and radiomic features such as shape and texture features were extracted based on contrast-enhanced T1-weighted imaging(CE-T1WI)images and subtracted images,and feature screening was performed using t-test and least absolute shrinkage and selection operator cross-validation,combined with the peak ratios of three metabolites of MRS,and a random forest algorithm was used to construct a high-grade glioma grading discrimination model and evaluate the model performance.Results:The area under the curve(AUC)of the model constructed based on CE-T1WI images was 0.78 in the test set;the AUC of the model constructed based on subtracted images was 0.81 in the test set;the AUC of the model constructed based on metabolite peak ratios from MRS was 0.80 in the test set;and the AUC of the model constructed based on CE-T1WI images and three metabolite peak ratios from MRS was 0.95 in the test set.Conclusions:Radiomics based on CE-T1WI images,subtraction images and MRS have good performance in identifying both gradeⅢand gradeⅣgliomas,with the subtraction model performing best in the single sequence model and the CE-T1WI combined with MRS model performing best in the combined sequence model.Multimodal radiomic analysis can provide a useful clinical aid for identifying gradeⅢandⅣgliomas.