AI-Based Approaches for Automatic Prescreening of Parkinson’s Disease Using EEG Signal during Sleep

Parkinson's disease (PD) is the second most common neurodegenerative disorder, and its prevalence is steadily increasing worldwide.However, diagnosing PD through clinical examination remains challenging and may be complicated by clinician subjectivity.In this study, we proposed an artificial intelligence-based method for the automatic prescreening of PD using nocturnal electroencephalography (EEG) signals recorded during various sleep stages.Patients with PD often exhibit rapid eye movement (REM) sleep behavior disorders that can manifest as abnormal EEG patterns during the REM stage.Therefore, we hypothesized that there would be detectable differences in sleep-state EEG patterns between patients with PD and healthy controls.Nocturnal EEG data were recorded during polysomnography in 65 participants.This study utilized pre-trained models followed by either random forest (TL-RF) or fully connected (TL-FC) layers.The results showed superior performance in the REM stage, with the TL-RF and TL-FC achieving F1-scores of 0.79 and TL-FC model attaining an F1-score of 0.83, respectively.In conclusion, we demonstrated the effectiveness of transfer learning-based models and the potential of REM-stage EEG features in assisting the diagnosis of PD.