Evaluating Effects of Resting-State Electroencephalography Data Pre-Processing on a Machine Learning Task for...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Resting-state electroencephalography (EEG) is a key tool for Parkinson's disease (PD) research.
• Machine learning (ML) models are increasingly used for PD classification based on EEG data.
• Variability in EEG preprocessing methods complicates the interpretation and reproducibility of ML-based PD studies.

Purpose of the Study:

• To investigate the impact of different resting-state EEG preprocessing stages on ML-based Parkinson's disease classification.
• To evaluate the influence of specific frequency bands (alpha and theta) and regions of interest on classification performance.
• To identify optimal preprocessing strategies for enhancing the accuracy of EEG-driven PD detection.

Main Methods:

• Utilized three distinct EEG datasets, applying four different preprocessing levels to each.
• Extracted power features from six predefined regions of interest.
• Employed Random Forest Classifiers for feature selection and Support Vector Machines for final classification.

Main Results:

• Muscle artifacts were found to artificially inflate evaluation metrics, highlighting the importance of effective artifact removal.
• Features from the alpha and theta frequency bands demonstrated superior performance when data underwent complete preprocessing.
• The extent of preprocessing significantly altered the classification outcomes across all datasets.

Conclusions:

• Thorough preprocessing of resting-state EEG data is crucial for reliable machine learning-based Parkinson's disease classification.
• Focusing on alpha and theta band features after comprehensive preprocessing offers the most promising results.
• Standardizing EEG preprocessing pipelines is essential for improving the consistency and validity of future PD research.