Department of Radiology, University of Iowa, Iowa City, Iowa, United States of America; Department of Biomedical Engineering, University of Iowa, Iowa City, Iowa, United States of America.
Department of Radiology, University of Iowa, Iowa City, Iowa, United States of America.
J Neuroradiol. 2019 May;46(3):179-185. doi: 10.1016/j.neurad.2018.05.006. Epub 2018 Jun 27.
This study explores whether objective, quantitative radiomic biomarkers derived from magnetic resonance (MR), positron emission tomography (PET), and computed tomography (CT) may be useful in reliably distinguishing malignant peripheral nerve sheath tumors (MPNST) from benign plexiform neurofibromas (PN).
A registration and segmentation pipeline was established using a cohort of NF1 patients with histopathological diagnosis of PN or MPNST, and medical imaging of the PN including MR and PET-CT. The corrected MR datasets were registered to the corresponding PET-CT via landmark-based registration. PET standard-uptake value (SUV) thresholds were used to guide segmentation of volumes of interest: MPNST-associated PET-hot regions (SUV≥3.5) and PN-associated PET-elevated regions (2.0<SUV<3.5). Quantitative imaging features were extracted from the MR, PET, and CT data and compared for statistical differences. Intensity histogram features included (mean, media, maximum, variance, full width at half maximum, entropy, kurtosis, and skewness), while image texture was quantified using Law's texture energy measures, grey-level co-occurrence matrices, and neighborhood grey-tone difference matrices.
For each of the 20 NF1 subjects, a total of 320 features were extracted from the image data. Feature reduction and statistical testing identified 9 independent radiomic biomarkers from the MR data (4 intensity and 5 texture) and 4 PET (2 intensity and 2 texture) were different between the PET-hot versus PET-elevated volumes of interest.
Our data suggests imaging features can be used to distinguish malignancy in NF1-realted tumors, which could improve MPNST risk assessment and positively impact clinical management of NF1 patients.
本研究旨在探讨从磁共振（MR）、正电子发射断层扫描（PET）和计算机断层扫描（CT）中提取的客观、定量的放射组学标志物是否可用于可靠地区分恶性外周神经鞘瘤（MPNST）与良性丛状神经纤维瘤（PN）。
通过建立一个 NF1 患者队列，对其进行组织病理学诊断为 PN 或 MPNST 的神经纤维瘤以及包括 MR 和 PET-CT 的 PN 医学成像进行配准和分割。通过基于标志的配准将校正后的 MR 数据集配准到相应的 PET-CT。使用 PET 标准摄取值（SUV）阈值来指导感兴趣区的分割：MPNST 相关的 PET 热点区域（SUV≥3.5）和 PN 相关的 PET 升高区域（2.0<SUV<3.5）。从 MR、PET 和 CT 数据中提取定量成像特征，并进行统计差异比较。强度直方图特征包括（均值、中值、最大值、方差、半高全宽、熵、峰度和偏度），而图像纹理则使用 Law 的纹理能量测度、灰度共生矩阵和邻域灰度差矩阵进行量化。
对于每个 NF1 患者，从图像数据中总共提取了 320 个特征。特征减少和统计检验从 MR 数据中确定了 9 个独立的放射组学标志物（4 个强度和 5 个纹理），从 PET 数据中确定了 4 个（2 个强度和 2 个纹理），这些标志物在 PET 热点与 PET 升高的感兴趣区之间存在差异。
我们的数据表明，成像特征可用于区分 NF1 相关肿瘤的恶性程度，这可能会改善 MPNST 的风险评估，并对 NF1 患者的临床管理产生积极影响。
