MRI-based differentiation of Parkinson's disease by cerebellar gray matter volume

Abstract

Background

The underlying mechanism of Parkinson's disease (PD) is associated with the neurodegeneration of the dopaminergic neurons, and the cerebellum plays a significant role together in non-motor and motor functions in PD progression. Morphological changes in the cerebellum can greatly impact patients' clinical symptoms, especially motor control symptoms, and may also help distinguish patients from healthy subjects. This study aimed to explore the potential of cerebellar gray matter volume, related to motor control function, as a neuroimaging biomarker to classify patients with PD and healthy controls (HC) by using voxel-based morphometric (VBM) measurements and support vector machine (SVM) methods based on independent component analysis (ICA).

Methods

Cerebellar gray matter volume was measured using VBM in patients with PD (n = 27) and HC (n = 16) from the Neurocon dataset. ICA analysis was performed on the gray matter volume of all subregions, resulting in 7 independent components. These independent components were then utilized for correlation analysis with clinical scales and trained as input features for the SVM model. PD patients (n = 20) and HC (n = 20) from the TaoWu dataset were used as test data to validate our SVM model.

Results

Among patients with PD, 3 out of the 7 independent components showed a significant correlation with clinical scales. The SVM model achieved an accuracy of 86 % in classifying PD patients and HC, with a sensitivity of 72.2 %, specificity of 88 %, and F1 Score of 76.5 %. The accuracy of the SVM model verification analysis using the TaoWu dataset was 70 %, with a sensitivity of 62.5 %, a specificity of 100 %, and the F1 Score was 76.9 %.

Conclusions

The results suggest that abnormal cerebellar gray matter volume, which is highly correlated with motor control function in Parkinson's patients, may serve as a valuable neuroimaging biomarker capable of distinguishing Parkinson's patients from healthy individuals. We observed that the combination of the ICA method and the SVM method produced an improved classification model. This model may function as an early warning tool that enables clinicians to conduct preliminary identification and intervention for patients with PD.