Abstract: Objective To investigate the classification of Parkinson’s disease (PD) through analyzing the profile of resting-state functional magnetic resonance of PD patients and healthy subjects. MethodsA total of 32 healthy subjects and 32 PD patients underwent resting-state functional magnetic resonance by 3.0T MRI. The resultant images were pre-treated by DPARSF software, yielding the amplitude of low-frequency fluctuation (ALFF), fractional amplitude of low-frequency fluctuation (fALFF) and regional homogeneity (ReHo) parameter maps. Then, the two-sample test was used to extract voxel values of significantly different brain regions which were viewed as the imaging features and the support vector machine classifier was used to train the model with the features. Finally, the classification accuracy, receiver operating characteristics (ROC) curve and the area under the curve (AUC) were evaluated. ResultsA total of 1167, 1678 and 2780 voxel values in abnormal brain regions were extracted from ALFF, fALFF and ReHo parameter maps. The highest classification accuracy was 96.97%, while the AUC value was 0.99. ConclusionsPD classification can be higher accurate through extracting the voxel values of abnormal brain regions from resting-state functional magnetic resonance images, providing an auxiliary basis for clinical diagnosis.