Radiodiagnostic Unit, université catholique de Louvain, Saint-Luc University Hospital, avenue Hippocrate 10, 1200 Brussels, Belgium.
J Neuroradiol. 2012 Dec;39(5):301-7. doi: 10.1016/j.neurad.2011.11.002. Epub 2011 Dec 22.
To assess the performance of parameters used in conventional magnetic resonance imaging (MRI), perfusion-weighted MR imaging (PWI) and visual texture analysis, alone and in combination, to differentiate a single brain metastasis (MET) from glioblastoma multiforme (GBM).
In a retrospective study of 50 patients (41 GBM and 14 MET) who underwent T2/FLAIR/T1(post-contrast) imaging and PWI, morphological (circularity, surface area), perfusion (rCBV in the ring-like tumor area, rCBV in the peritumoral area, percentage of signal intensity recovery at the end of first pass) and texture parameters in the peritumoral area were estimated. Statistical differences and performances were assessed using Wilcoxon's test and receiver operating characteristic curves, respectively. Multiparametric classification of tumors was performed using k-means clustering.
Significant statistical differences in circularity, surface area, rCBVs, percentage of signal intensity recovery and texture parameters (energy, entropy, homogeneity, correlation, inverse differential moment, sum average) were observed between MET and GBM (P<0.05). Moderate-to-good classification performances were found with these parameters. Clustering based on rCBV and texture parameters (contrast, sum average) differentiated MET from GBM with a sensitivity of 92% and a specificity of 71%.
Combining perfusion and visual texture parameters within a statistical classifier significantly improved the differentiation of a single brain MET and GBM.
评估常规磁共振成像（MRI）、灌注加权磁共振成像（PWI）和视觉纹理分析中使用的参数的性能，单独或联合使用这些参数来区分单个脑转移瘤（MET）和胶质母细胞瘤（GBM）。
在一项回顾性研究中，对 50 名患者（41 名 GBM 和 14 名 MET）进行了 T2/FLAIR/T1（增强后）成像和 PWI 检查，评估了形态学（圆形度、表面积）、灌注（环形肿瘤区域的相对脑血容量 [rCBV]、肿瘤周围区域的 rCBV、初次通过结束时的信号强度恢复百分比）和肿瘤周围区域的纹理参数。使用 Wilcoxon 检验和受试者工作特征曲线分别评估统计差异和性能。使用 k-均值聚类对肿瘤进行多参数分类。
MET 和 GBM 之间在圆形度、表面积、rCBVs、信号强度恢复百分比和纹理参数（能量、熵、均匀性、相关性、逆差分矩、总和平均）方面存在显著统计学差异（P<0.05）。这些参数具有中等至良好的分类性能。基于 rCBV 和纹理参数（对比度、总和平均）的聚类可以区分 MET 和 GBM，敏感性为 92%，特异性为 71%。
在统计分类器中结合灌注和视觉纹理参数可以显著提高单个脑 MET 和 GBM 的鉴别能力。
