EEGMMIDB Permutation-Entropy (PE) Decoding Pipeline

Pipeline used to benchmark permutation-entropy features on the EEGMMIDB. It decodes motor execution (ME) and motor imagery (MI) tasks using:

• Temporal PE
• HV spatial PE (horizontal/vertical triplets)
• Cross-like spatial PE (plus-shaped neighboring electrodes)
• Spatiotemporal PE (center, up, right, and left electrodes combined with a time-lagged center electrode signal)
• Combined PE feature models, plus ablations and sanity checks (within-pair label permutation and capacity-matched controls)

What this repository does

• Loads EEGMMIDB EDF files (subjects S001–S109)
• Preprocesses EEG (EEG-only, resampling, average reference, optional notch)
• Extracts paired baseline/task epochs per cue
• Applies band-invariant artifact rejection and trial capping
• Computes PE features under two scopes: whole scalp and sensorimotor ROI
• Runs within-subject decoding using group-aware CV (trial-pair grouped)
• Writes per-subject and group summaries, plus figures

Requirements

• Python 3.9+ is recommended
• Dependencies: see requirements.txt

Install:

pip install -r requirements.txt

Data

You can download EEGMMIDB (PhysioNet MI/ME dataset) and then place it in:

./data/EEGMMIDB/S001
./data/EEGMMIDB/S002
./data/EEGMMIDB/S003
...

Configuration

You can copy the example configuration and then edit paths as needed:

cp config.example.json config.json

Important settings:

• DATA_ROOT: path containing S001, S002, S003, ...
• OUT_FOLDER: where outputs are written (default: ./outputs/...)
• BANDS, epoch windows, rejection thresholds, CV folds, etc. (see run_pipeline.py)

The data path can also be set through the environment variable:

export EEGMMIDB_ROOT="/path/to/EEGMMIDB"

To run the code

python run_pipeline.py

Outputs

A run folder is created under OUT_FOLDER and will contain files such as:

• run_config.txt, run_config.json
• subject_summary.csv
• decoding_by_subject.csv
• decoding_group_summary.csv
• family_comparisons_stats.csv
• feature_condition_summary.csv
• ablation_by_subject.csv, ablation_group_summary.csv
• permutation_stats.csv
• matched_subjects_MI_ME_summary.csv
• cross_count_distribution_summary.csv
• capacity_matched_controls.csv
• Figures: Fig_decoding_bar_with_dots_<band>_<scope>.png

Other notes

• This pipeline uses deterministic RNG (RNG_SEED)
• This pipeline uses group-aware CV (trial-pair grouping) when available (StratifiedGroupKFold)