Unraveling Parkinson's disease motor subtypes: A deep learning approach based on spatiotemporal dynamics of EEG microstates

Background: Despite prior studies on early-stage Parkinson's disease (PD) brain connectivity and temporal patterns, differences between tremor-dominant (TD) and postural instability/gait difficulty (PIGD) motor subtypes remain poorly understood. Our study aims to understand the contribution of altered brain network dynamics to heterogeneous motor phenotypes in PD for improving personalized treatment. Methods: Electroencephalography (EEG) microstate dynamics were firstly used to capture spatiotemporal brain network changes. A deep learning model was developed to classify PD motor subtypes where spatial variability and electrode location data were incorporated into the analysis. Results: Compared to healthy individuals, both PD-TD and PD-PIGD patients showed increased local segregation of brain regions. The PD-PIGD subtype had more severe and extensive disorganization in microstate A dynamics, suggesting greater disruption in auditory and motor-related networks. Incorporating spatial information significantly improved the accuracy of subtype classification, with an AUC of 0.972, indicating that EEG microstate dynamic spatial patterns reflect distinct PD motor pathologies. The increased spatial variability in the PD-PIGD group was more closely associated with motor impairments. Conclusions: This study presents a novel framework for differentiating PD motor subtypes and emphasizes dynamic brain network features as potential markers for understanding motor symptom variability in PD, which may contribute to the development of personalized treatment strategies. Trial registration: ChiCTR2300067657.