Yang Yunjun, Xu Zhenyu, Li Cheng, Wang Chengming, Zhao Hai, Xu Zhifeng
Department of Radiology, The First People's Hospital of Foshan, #81 North Lingnan Avenue, Foshan, Guangdong, China.
Neurol Sci. 2025 May;46(5):2103-2113. doi: 10.1007/s10072-024-07956-0. Epub 2024 Dec 26.
Identifying Parkinson's disease (PD) during its initial phases presents considerable hurdles for clinicians.
To examine the feasibility and efficacy of a machine learning model based on quantitative multiparametric magnetic resonance imaging (MRI) features in identifying early-stage PD.
We recruited 33 participants, including 19 with early-stage PD, 14 with advanced-stage PD and 20 healthy control subjects. Each participant underwent both quantitative susceptibility mapping (QSM) and diffusion kurtosis imaging (DKI). We utilized combined QSM and DKI features to establish a support vector machine (SVM) model to identify early-stage PD.
When comparing early-stage PD with healthy controls, the SVM model exhibited moderate performance, achieving a training set accuracy of 0.78 and an area under the receiver operating characteristic curve (AUC) of 0.90, and the accuracy of 0.77 (AUC = 0.87) in the test set. When comparing advanced-stage PD with healthy controls, the SVM model exhibited equally high accuracy in both training (0.97, AUC = 0.97) and test (0.94, AUC = 0.94) sets. In discriminating between early-stage PD and advanced-stage PD, the SVM model achieved an accuracy of 0.80 (AUC = 0.81) in the training set and an accuracy of 0.71 (AUC = 0.72) in the test set. The mean kurtosis feature of DKI in the substantia nigra, played a significant role in classification.
These findings suggest that early PD is associated with specific MRI features reflecting magnetic susceptibility and microstructural changes. The SVM model combining quantitative QSM and DKI features holds promise for improving early PD diagnosis.
在帕金森病（PD）的初始阶段识别该病对临床医生而言存在诸多重大障碍。
研究基于定量多参数磁共振成像（MRI）特征的机器学习模型在识别早期PD中的可行性和有效性。
我们招募了33名参与者，包括19名早期PD患者、14名晚期PD患者和20名健康对照者。每位参与者均接受了定量磁化率成像（QSM）和扩散峰度成像（DKI）。我们利用QSM和DKI的联合特征建立了一个支持向量机（SVM）模型来识别早期PD。
将早期PD与健康对照进行比较时，SVM模型表现出中等性能，训练集准确率达到0.78，受试者操作特征曲线（AUC）下面积为0.90，测试集准确率为0.77（AUC = 0.87）。将晚期PD与健康对照进行比较时，SVM模型在训练集（0.97，AUC = 0.97）和测试集（0.94，AUC = 0.94）中均表现出同样高的准确率。在区分早期PD和晚期PD时，SVM模型在训练集中的准确率为0.80（AUC = 0.81），在测试集中的准确率为0.71（AUC = 0.72）。黑质中DKI的平均峰度特征在分类中起重要作用。
这些发现表明，早期PD与反映磁化率和微观结构变化的特定MRI特征相关。结合定量QSM和DKI特征的SVM模型有望改善早期PD的诊断。
