Department of Radiology, The First Affiliated Hospital, Nanchang University, 17 Yongwaizheng Street, Nanchang, Jiangxi 330006, China.
Department of Radiology and Tianjin Key Laboratory of Functional Imaging, Tianjin Medical University General Hospital, Tianjin 300052, China.
Mult Scler Relat Disord. 2022 May;61:103756. doi: 10.1016/j.msard.2022.103756. Epub 2022 Mar 23.
Ischemic vasculopathy, particularly small-vessel disease, may mimic multiple sclerosis (MS) located in the periventricular or subcortical region on magnetic resonance (MR) examinations and should be included in the differential diagnosis of MS-like lesions.
To evaluate the performance of a T2-weighted imaging (T2WI)-based radiomic signature to distinguish MS lesions from lesions corresponding to ischemic demyelination, which often mimics MS on MRI.
A retrospective study was conducted in 38 patients (627 lesions) with MS and 914 patients (2466 lesions) with lesions mimicking ischemic demyelination in the periventricular or subcortical region. All patients underwent 3 T MRI. A total of 472 radiomic features were extracted from the T2WI data of each patient. Intraclass correlation coefficients were used to select the features with excellent stability and repeatability. Then, we used the minimum-redundancy maximum-relevance (mRMR) and least absolute shrinkage and selection operator (LASSO) algorithms for feature selection. After feature selection, various classifiers (including logistic regression, decision tree, AdaBoost, random forest (RF), or support vector machine (SVM)) were trained. The performance of each classifier was validated in the test set by determining the area under the curve (AUC).
Nine features were selected to distinguish MS lesions from the similar lesions of ischemic demyelination. The radiomic signature showed a significant difference between the MS and ischemic demyelination patients (p < 0.01). RF and SVM were overfitted. The LASSO logistic regression model was the best-performing radiomic model,with an AUC, accuracy, sensitivity, and specificity of 0.900 (95% CI: 0.883-0.918), 87.0%, 58.9% and 95.2%, respectively, in the training set and 0.828 (95% CI: 0.791-0.864), 87.7%, 53.6% and 94.4%, respectively, in the validation set.
The T2WI-based radiomic signature can effectively differentiate MS patients from patients with MS-like lesions due to ischemic demyelination.
缺血性血管病，特别是小血管疾病，可能在磁共振成像（MR）检查中模仿多发性硬化症（MS）位于脑室周围或皮质下区域，应包括在 MS 样病变的鉴别诊断中。
评估基于 T2 加权成像（T2WI）的放射组学特征在区分 MS 病变与缺血性脱髓鞘病变方面的性能，后者在 MRI 上常模仿 MS。
对 38 例 MS 患者（627 个病变）和 914 例脑室周围或皮质下区域具有缺血性脱髓鞘病变模拟病变的患者（2466 个病变）进行回顾性研究。所有患者均接受 3 T MRI 检查。从每位患者的 T2WI 数据中提取了 472 个放射组学特征。采用组内相关系数（intraclass correlation coefficient）选择稳定性和可重复性好的特征。然后，我们使用最小冗余最大相关性（minimum-redundancy maximum-relevance，mRMR）和最小绝对值收缩和选择算子（least absolute shrinkage and selection operator，LASSO）算法进行特征选择。特征选择后，使用逻辑回归、决策树、AdaBoost、随机森林（random forest，RF）或支持向量机（support vector machine，SVM）等多种分类器进行训练。通过确定曲线下面积（area under the curve，AUC），在测试集中验证每个分类器的性能。
选择了 9 个特征来区分 MS 病变和缺血性脱髓鞘病变。放射组学特征在 MS 患者和缺血性脱髓鞘患者之间有显著差异（p < 0.01）。RF 和 SVM 存在过拟合现象。LASSO 逻辑回归模型是表现最佳的放射组学模型，在训练集和验证集的 AUC、准确率、敏感度和特异度分别为 0.900（95%CI：0.883-0.918）、87.0%、58.9%和 95.2%、0.828（95%CI：0.791-0.864）、87.7%、53.6%和 94.4%。
基于 T2WI 的放射组学特征可有效区分 MS 患者与缺血性脱髓鞘引起的 MS 样病变患者。
