Parkinson's disease tremor prediction using EEG data analysis-A preliminary and feasibility study

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Resting tremor is a key symptom of Parkinson's disease (PD) often unresponsive to medication.
• Predicting tremor onset is crucial for optimizing interventions like deep brain stimulation.
• Current prediction methods require further development for clinical application.

Purpose of the Study:

• To introduce a novel methodology for predicting resting tremors in PD patients.
• To utilize electroencephalogram (EEG) time-series data for tremor prediction.
• To identify discriminative EEG features for pre-tremor condition detection.

Main Methods:

• A machine learning approach was developed to predict PD hand tremors from EEG.
• Statistical analyses and post-hoc tests identified key EEG features (e.g., form factor).
• A K-Nearest Neighbors (KNN) classifier was trained using selected features from limited EEG channels and bands.

Main Results:

• Specific EEG features and channels (F3, F7, P4, CP2, FC6, C4) and bands (Delta, Gamma) were found to be most discriminative.
• The KNN classifier achieved a pre-tremor prediction accuracy of 73.67%.
• Accurate tremor prediction was feasible using a limited set of EEG data.

Conclusions:

• This study demonstrates the feasibility of using EEG time-series for predicting PD hand tremors.
• It is the first study to show the predictive capability of EEG for PD tremors.
• Further research with extended data and diverse brain dynamics is necessary for clinical implementation.