颅内胶质瘤MRI影像组学评估颅内高级别与低级别胶质瘤

目的:探讨MRI影像组学在颅内胶质瘤术前分级评估中的应用价值,提高MRI对颅内胶质瘤的诊断效能.方法:回顾分析58例经病理证实的颅内胶质瘤患者的MRI资料,使用Matlab软件提取病灶特征并使用Lasso软件(R语言3.4.0版)进行降维处理,降维后的特征再进行Logistic多元回归相关性分析,从而筛选出具有诊断价值且与肿瘤级别显著相关的影像组学特征,运用受试者工作特征(ROC)曲线对特征参数的诊断效能进行分析.结果:DWI序列降维后经Logistic多元回归分析筛选出影像组学相关特征为:均值(P=0.04)、RMS(P=0.04)、Percentile 65(P=0.023)、Percentile-70(P=0.03)和Percentile-75 (P=0.009).DWI提取特征中Percentile-75对高、低级别胶质瘤的诊断效能最高,ROC曲线下面积(AUC)为0.793.T2 WI序列及T1 WI增强序列未筛选出与肿瘤分级相关的参数.在热力图上DWI序列Percentile-75特征与肿瘤分级的相关性较明显.结论:在T2 WI、DWI和T1WI增强三个序列中,DWI序列对高低级别胶质瘤的鉴别效能较高,该序列上所筛选的特征参数中以Percentile-75的诊断价值最高.

Diagnosis efficacy of MRI radiomics in grading of cerebral glioma before operation

Objective:Discussing the application of radiomics was used in MRI image by diagnosis cerebral glioma and improving the diagnosis efficacy.Methods:The research findings 58 cases of cerebral glioma by pathology were analyzed retrospectively.Characteristics were picked up from tumor in two dimensional planes by Matlab software,and the key factors were retrieved by Lasso package in R-software with 3.4.0 generation.Stepwise multiple logistic regression analysis was used to determin the independent factor that MRI image can diagnosis grading of cerebral glioma.Using box-type and scat ter to describe the factor and at the same time using the ROC curve to analysis the key factor diagnosis efficacy.Results:The key factors of the lesions on MRDWI were analyzed by stepwise multiple logistic regression as the best select examination.The correlative factor as following:mean value (P =0.04),RMS (P=0.04),percentile-65 (P =0.023),percentile-70 (P =0.03) and Percentile-75 (P=0.009).The factor of percentile-75 as the dependent correlative factor and it is the highest diagnosis efficacy with the AUC was 0.793.Characteristics of Percentile-75 have high correlation with pathology grading where filling in more area red colore in hotmap of DWI.T1 WI with contrast-enhanced and T2 WI had not correlation factors.Conclusions:DWI was the highest diagnosis efficacy sequence among T1 WI contrast-enhanced with T2 WI,and the factor that percentile-75 was the independent correlation factor with the best diagnosis efficacy.