EEG Sensor-Based Parkinson's Disease Detection Using a Multi-Domain Feature Fusion Network.

Area of Science:

• Neuroscience and Biomedical Engineering
• Signal Processing and Machine Learning

Background:

• Parkinson's disease (PD) diagnosis relies on accurate identification, with electroencephalography (EEG) offering practical real-time brain signal acquisition.
• Traditional single-domain analysis of non-stationary EEG signals is insufficient for robust Parkinson's disease feature extraction.

Purpose of the Study:

• To develop and evaluate a novel multi-domain feature fusion model for enhanced Parkinson's disease detection using EEG.
• To investigate the efficacy of integrating temporal, frequency-domain, and wavelet-domain EEG features for improved classification accuracy.

Main Methods:

• Proposed the Multi-Domain Fusion Network (MDF-Net), integrating temporal, frequency, and wavelet domains for EEG classification.
• Utilized a Temporal Attention-enhanced Temporal Convolutional Network (TTCN) for temporal dependency capture and a 1D Convolutional Neural Network mixer (Cmix) for multi-channel feature fusion.
• Constructed and analyzed an EEG dataset comprising 415 subjects (126 PD patients, 289 controls) using 5-fold cross-validation.

Main Results:

• MDF-Net achieved a classification accuracy of 92.3%, an F1-score of 87.3%, and an Area Under the Curve (AUC) of 0.943.
• Demonstrated that multi-domain feature fusion significantly enhances Parkinson's disease detection performance compared to single-domain methods.
• EEG sensor-based analysis shows strong potential for practical clinical application in objective PD diagnosis.

Conclusions:

• The proposed MDF-Net effectively leverages multi-domain feature fusion for accurate Parkinson's disease identification from EEG signals.
• This study provides a valuable methodological reference for developing objective, practical computer-aided diagnostic tools for Parkinson's disease.