EEG Classification for Multiple Sclerosis (MS) Detection

This project explores the feasibility of classifying Multiple Sclerosis (MS) patients using 5-minute resting-state EEG data. The goal was to build a machine learning pipeline capable of distinguishing between MS patients and healthy controls based on frequency-based analysis.

Project Overview

The study involved a small-scale dataset of 10 subjects:

• 5 MS Patients

• 5 Healthy Controls

Given the limited number of subjects, the signals were windowed to increase the number of samples available for training and testing.

Methodology

1. Signal Processing & Feature Extraction

The raw EEG data was processed by calculating the Band Power for every channel across specific frequency ranges:

• Delta: 0.5 – 4 Hz

• Theta: 4 – 8 Hz

• Alpha: 8 – 16 Hz

• Beta: 16 – 31 Hz

• Gamma: 31 – 50 Hz

Note

The signal processing method used in this study is the same as the signal processing method used in this study by Sasmaz et al¹.

2. Machine Learning Pipeline

We implemented and compared several classification algorithms to identify the best performer for this specific signal type:

• LDA (Linear Discriminant Analysis)

• SVM (Support Vector Machine)

• Random Forest

• Logistic Regression

Data Split

• 80% of the windowed data was used for training.

• 20% of the data was reserved for testing and validation.

File Structure

File Description

• oop_channel.py Handles data ingestion, windowing logic, and the calculation of frequency band powers.
• MS_eeg_detection_model.ipynb Contains the machine learning implementation, including model training, the 80/20 split, and performance evaluation. Requirements

Important

To run these scripts, you will need the following Python libraries:

• numpy

• pandas

• scikit-learn

• scipy (for signal processing)

Note

This research was ultimately discontinued due to an insufficient amount of data to achieve statistically significant or generalizable results. It remains a proof-of concept for the processing pipeline.

Conclusion

While the machine learning models were successfully implemented, the sample size (10 subjects) proved too small for a robust clinical classification model. Future iterations of this study would require a significantly larger dataset to validate the efficacy of these EEG biomarkers.

Footnotes

1. Şaşmaz Karacan, S., & Saraoğlu, H. M. (2024). A simplified method for relapsing-remitting multiple sclerosis detection: Insights from resting EEG signals. Computers in Biology and Medicine, 178, 108728. https://doi.org/10.1016/j.compbiomed.2024.108728 ↩