Abnormal Subthalamic Nucleus Functional Connectivity and Machine Learning Classification in Parkinson's Disease: A Multisite Functional Magnetic Resonance Imaging Study

Bin Qin^1,2Yisi Tang^1,3Huixun Qin^1,3Wen Gao^1,3Shusheng Liao^1,3Mingxiu Yang^1,3*

• ¹Department of Neurology, Liuzhou People's Hospital, Liuzhou, China
• ²Department of Neurology, The First Affiliated Hospital of Guangxi Medical University, Nanning, China
• ³Liuzhou Key Laboratory of Neurointervention, Liuzhou, China

Introduction: Parkinson’s disease (PD) is a progressive neurodegenerative disorder imposing a significant global burden, characterized by motor dysfunction linked to aberrant basal ganglia activity. This multisite study leveraged pooled resting-state functional magnetic resonance imaging (rs-fMRI) data to characterize subthalamic nucleus (STN) functional connectivity (FC) abnormalities and evaluate their utility in machine learning classification of PD. Methods: We analyzed rs-fMRI data from 232 participants (158 PD, 74 healthy controls [HCs]) across four repositories (PPMI, OpenfMRI, FCP/INDI). Seed-based FC analysis focused on bilateral STNs. Group comparisons (PD vs. HCs) employed two-sample t-tests with Gaussian Random Field correction (GRF). Support Vector Machine (SVM) classifier, incorporating significant FC features, was used for diagnostic classification. Results: Patients with PD exhibited significant bilateral reductions in STN FC compared to HCs. Specifically, left STN showed decreased connectivity with the left superior temporal gyrus and right supramarginal gyrus, while right STN showed decreased connectivity with the right superior temporal gyrus, left middle temporal gyrus, and left inferior frontal gyrus (voxel P<0.005, cluster P<0.05, GRF corrected). The SVM classifier utilizing these FC features achieved high diagnostic accuracy (89.1%), sensitivity (97.7%), specificity (75.8%), and an area under the receiver operating characteristic curve of 0.931 in the validation set. Conclusion: This study suggests that STN-temporal/parietal hypoconnectivity should be further investigated as a possible core feature of PD. Furthermore, it demonstrates the high translational potential of STN-centric FC patterns as diagnostic biomarkers when integrated with machine learning, paving the way for improved PD classification and future applications in personalized neuromodulation strategies.