This project focuses on preprocessing and cleaning scalp EEG signals from the CHB-MIT Scalp EEG Database. The primary objective is to reduce ocular (EOG) and muscle (EMG) artifacts in order to obtain cleaner neural signals suitable for downstream analysis (e.g., seizure-related studies).
The pipeline is based on Independent Component Analysis (ICA) to decompose EEG recordings into independent sources and suppress artifact-related components.
- Preprocessing: Channel normalization and band-pass filtering using MNE-Python.
- Artifact Removal: ICA-based removal of eye-blink and muscle-related noise.
- Signal Assessment: Power Spectral Density (PSD) analysis for qualitative evaluation of frequency-band preservation.
- Quality Metrics: Computation of variance reduction percentages as a proxy for artifact suppression strength.
The project uses the CHB-MIT Scalp EEG Database collected at Boston Children's Hospital.
- Subjects: 24 patients (chb01–chb24)
- Format: EDF (European Data Format)
Note: The dataset files are not included in this repository due to size and licensing constraints.
- Loading raw EEG recordings from EDF files.
- Applying band-pass (typically 1–40 Hz) and notch (60 Hz) filters to reduce noise.
- Performing ICA decomposition to separate independent signal sources.
- Identifying artifact-related components using variance-based heuristics.
- Reconstructing cleaned EEG signals after component rejection.
The notebook produces a Global_Quality_Report.csv summarizing artifact reduction statistics across all processed recordings. Visualizations illustrate the distribution of variance reduction values.
Higher variance reduction indicates successful suppression of strong artifacts, while values close to zero typically correspond to clean or low-activity (e.g., sleep) segments.
Python 3.x
MNE-Python
NumPy
Pandas
Matplotlib
Seaborn
SciPy
This project is released for academic and research purposes.