摘要
目的肿瘤亚区影像组学特征结合分类算法在脑胶质母细胞瘤MGMT启动子甲基化状态中的应用价值。方法回顾性对术后病理证实的脑胶母质瘤患者进行分析,经过排除纳入标准,最终40例患者纳入研究。调取患者的MRI图像及病理结果,在T_(1)WI增强序列分别对图像各感兴趣区域进行勾画,包括强化区、非强化区、整体肿瘤,利用相关软件捕捉每个病灶的形状、质地和边缘锐度的定量图像特征进行降维后分析,探讨影像组学在胶质母细胞瘤MGMT启动子甲基化状态的应用方法及诊断效能。结果基于T_(1)增强扫描的各感兴趣区的某种组学特征与胶质母细胞瘤MGMT甲基化启动子阳性具有一定的相关性,非强化区组学特征相关性最好,7种机器学习模型中随机森林分类器表现出了很好的“鲁棒性”,测试集准确性达到了72%,分层10折交叉验证的曲线下面积达到0.73,特异度及敏感度分别达到0.65±0.39、0.82±0.23(P=0.003)。结论MRI T_(1)增强扫描的放射组学特征对胶质母细胞瘤MGMT启动子甲基化具有诊断价值;肿瘤亚区纹理特征研究为中枢神经系统胶质瘤影像组学研究提供新的方法与思路;结合分类算法可以较好预测MGMT启动子甲基化状态。
Objective To investigate the application value of tumor subregion radiomic features combined with classification algorithm in the evaluation of MGMT promoter methylation status in glioma.Methods Retrospective analysis was performed on patients who pathologically diagnosed as glioblastoma.In total,40 patients enrolled according to the inclusion and exclusion criteria.MRI and pathological data were obtained and ROIs including enhanced areas,non enhanced areas and whole tumor region were segmented manually on enhanced T_(1)WI.Quantitative image features of shape,texture and edge sharpness of each lesion were captured by relevant software,and dimensionality reduction analysis was conducted to explore the optimal method and diagnostic efficacy of radiomics in evaluation of MGMT promoter methylation status.Results Radiomic features from each region of interest based on T_(1) C were correlated with the positive methylation promoter status of MGMT in glioma,and the radiomic features from the non enhanced region had the best correlations.Among the seven machine learning models,RF showed good“robustness”.The accuracy in the test set was 72%,the AUC of the layered 10-fold cross validation was 0.73,the specificity and sensitivity were 0.65±0.39 and 0.82±0.23,respectively,P=0.003.Conclusion The radiomic characteristics of T_(1) enhanced MRI have diagnostic value for MGMT promoter methylation statues in glioblastoma.The study of tumor subregion texture features provides new methods and ideas for the study of central nervous system glioma imaging.Combined with the classification algorithm,the methylation status of MGMT promoter could be better predicted.
作者
王刚
赵晶晶
努尔比耶姆·阿布力克木
马晓山
姜春晖
丁爽
王云玲
贾文霄
王俭
WANG Gang;ZHAO Jingjing;Nuerbiyemu·Abulikemu(Imaging Center,the First Affiliated Hospital of Xinjiang Medical University,Urumqi,Xinjiang Uygur Autonomous Region 830000,P.R.China)
出处
《临床放射学杂志》
北大核心
2022年第7期1222-1226,共5页
Journal of Clinical Radiology
