Overview

This repository contains code and resources for a computational neuroscience research project focused on optimizing machine learning models for identifying chronic pain patients using resting-state electroencephalographic (EEG) signals. The primary objective was to improve upon previous methodologies that achieved a 57% accuracy, slightly surpassing chance levels, as reported in the study "Brain dysfunction in chronic pain patients assessed by resting-state electroencephalography", which can be found in https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7195856/.

Key Contributions

• Expansion of Feature Space: The research extended beyond the use of power spectral density (PSD) features in electrode space by computing average relative power features across diverse brain regions in source space.
• Significant Findings: Statistically significant differences in features within the insular cortex and postcentral sulcus were discovered between patient and control groups, enhancing the understanding of chronic pain markers in EEG signals.
• Enhanced Models: The project explored multiple machine learning models, including SVM, Adaboost, and Gradient Boost, resulting in a 7% improvement over baseline models.

Contents

• Code: Python scripts and Jupyter notebooks used for data preprocessing, feature extraction, model training, and evaluation.
• Data: The dataset can be found here: https://osf.io/srpbg/
• Results: Detailed results, including performance metrics, visualizations, and analysis outputs can be found in the jupyter notebooks

Experiments

EEG Analysis Baseline:

The baseline model includes five sets of features: the relative powers for all channels, all regions, statistically significant channels, all source labels, and statistically significant source labels.

EEG Analysis 1:

With 1/f noise reduction

EEG Analysis 2:

With 1/f noise as features

Getting Started

To replicate or build upon this research, follow these steps:

1. Download the public dataset
2. Preprocess the dataset
3. Follow the steps in the jupyter notebooks, which include feature extraction, model training, and evaluation.

References

Ta Dinh S, Nickel MM, Tiemann L, May ES, Heitmann H, Hohn VD, Edenharter G, Utpadel-Fischler D, Tölle TR, Sauseng P, Gross J, Ploner M. Brain dysfunction in chronic pain patients assessed by resting-state electroencephalography. Pain. 2019 Dec;160(12):2751-2765. doi: 10.1097/j.pain.0000000000001666. Erratum in: Pain. 2020 Jul 1;161(7):1684. PMID: 31356455; PMCID: PMC7195856.