EEG and EMG Signal Analysis for the Early Detection of Parkinson's Disease

Parkinson's Disease (PD) is a progressive neurodegenerative disorder that significantly impacts motor and cognitive function. Early and accurate diagnosis is a significant clinical challenge. This study proposes a hybrid deep learning framework that integrates Electroencephalography (EEG) and Electromyography (EMG) signals to classify PD patients. EEG signals were collected using the Emotiv Epoc X headset (14 channels, 10–20 system). At the same time, EMG data were acquired from three sensors placed according to the SENIAM standard (biceps, flexor carpi, extensor digitorum). Publicly available datasets, including the San Diego PD EEG dataset, were employed for model training and evaluation. Preprocessing included 1–50 Hz band pass filtering, Independent Component Analysis (ICA) for artifact removal, and epoch segmentation for EEG, while EMG signals underwent 20–450 Hz filtering, rectification, and RMS smoothing. A hybrid Convolutional Neural Network (CNN)– Long Short-Term Memory (LSTM) architecture was developed in Python to capture spatial and temporal dependencies in the multimodal bio signals. The model achieved 99% classification accuracy with a training loss of 0.14, demonstrating strong predictive power for early-stage PD detection. Despite promising results, the study is limited by the use of only 14 EEG electrodes and three EMG electrodes, with recordings restricted to rest conditions. Future work will expand electrode coverage, incorporate additional limb-based EMG sensors, and evaluate PD-related neural and muscular activity during more diverse tasks, such as puzzle solving, handwriting, and typing. This research highlights the potential of multimodal deep learning approaches for early and non-invasive PD diagnosis.