Parvaze P Suhail, Bhattacharjee Rupsa, Verma Yogesh Kumar, Singh Rakesh Kumar, Yadav Virendra, Singh Anup, Khanna Gaurav, Ahlawat Sunita, Trivedi Richa, Patir Rana, Vaishya Sandeep, Shah Tejas J, Gupta Rakesh K
Philips Innovation Campus, Bangalore, India.
Department of Radiology & Biomedical Imaging, University of California, San Francisco, California, USA.
NMR Biomed. 2023 May;36(5):e4884. doi: 10.1002/nbm.4884. Epub 2022 Dec 23.
The peritumoral vasogenic edema (PVE) in brain tumors exhibits varied characteristics. Brain metastasis (BM) and meningioma barely have tumor cells in PVE, while glioblastoma (GB) show tumor cell infiltration in most subjects. The purpose of this study was to investigate the PVE of these three pathologies using radiomics features in FLAIR images, with the hypothesis that the tumor cells might influence textural variation. Ex vivo experimentation of radiomics analysis of T1-weighted images of the culture medium with and without suspended tumor cells was also attempted to infer the possible influence of increasing tumor cells on radiomics features. This retrospective study involved magnetic resonance (MR) images acquired using a 3.0-T MR machine from 83 patients with 48 GB, 21 BM, and 14 meningioma. The 93 radiomics features were extracted from each subject's PVE mask from three pathologies using T1-dynamic contrast-enhanced MR imaging. Statistically significant (< 0.05, independent samples T-test) features were considered. Features maps were also computed for qualitative investigation. The same was carried out for T1-weighted cell line images but group comparison was carried out using one-way analysis of variance. Further, a random forest (RF)-based machine learning model was designed to classify the PVE of GB and BM. Texture-based variations, especially higher nonuniformity values, were observed in the PVE of GB. No significance was observed between BM and meningioma PVE. In cell line images, the culture medium had higher nonuniformity and was considerably reduced with increasing cell densities in four features. The RF model implemented with highly significant features provided improved area under the curve results. The possible infiltrative tumor cells in the PVE of the GB are likely influencing the texture values and are higher in comparison with BM PVE and may be of value in the differentiation of solitary metastasis from GB. However, the robustness of the features needs to be investigated with a larger cohort and across different scanners in the future.
脑肿瘤周围的血管源性水肿（PVE）表现出不同的特征。脑转移瘤（BM）和脑膜瘤在PVE中几乎没有肿瘤细胞，而胶质母细胞瘤（GB）在大多数病例中显示有肿瘤细胞浸润。本研究的目的是利用FLAIR图像中的放射组学特征研究这三种病变的PVE，假设肿瘤细胞可能影响纹理变化。还尝试对含有和不含悬浮肿瘤细胞的培养基的T1加权图像进行放射组学分析的体外实验，以推断肿瘤细胞增加对放射组学特征的可能影响。这项回顾性研究涉及使用3.0-T磁共振机器从83例患者获取的磁共振（MR）图像，其中包括48例GB、21例BM和14例脑膜瘤。使用T1动态对比增强MR成像从三种病变的每个受试者的PVE掩码中提取93个放射组学特征。考虑具有统计学意义（<0.05，独立样本T检验）的特征。还计算特征图用于定性研究。对T1加权细胞系图像进行了同样的操作，但使用单因素方差分析进行组间比较。此外，设计了一种基于随机森林（RF）的机器学习模型来对GB和BM的PVE进行分类。在GB的PVE中观察到基于纹理的变化，尤其是更高的不均匀性值。BM和脑膜瘤的PVE之间未观察到显著性差异。在细胞系图像中，培养基具有更高的不均匀性，并且在四个特征中随着细胞密度的增加而显著降低。用高度显著特征实施的RF模型提供了改进的曲线下面积结果。GB的PVE中可能的浸润性肿瘤细胞可能影响纹理值，并且与BM的PVE相比更高，可能在鉴别孤立性转移瘤与GB中具有价值。然而，这些特征的稳健性未来需要在更大的队列和不同的扫描仪上进行研究。
