EEG Framework

A modular Python framework for EEG-based Brain-Computer Interface (BCI) research, with a focus on Brain-to-Speech (BTS) and Motor Imagery (MI) classification. Built on top of MNE-Python and scikit-learn conventions.

---

Overview

The framework provides end-to-end support for EEG classification pipelines:

• Dataset readers
• Preprocessing
• Feature extraction
• Deep learning models
• Evaluation
• Plotting

---

Framework Structure

eeg_framework/
├── datasets/
├── preprocessing/
├── features/
├── models/
├── evaluate/
└── plotting/

---

Datasets

OpenBMI MI

54 subjects, 62 EEG channels, ~1000 Hz.
2 classes: left, right.

from eeg_framework.datasets import load_openbmi_mi

epochs_train = load_openbmi_mi(root_dir="data/OpenBMI MI", split="train", subjects=[1])
epochs_test  = load_openbmi_mi(root_dir="data/OpenBMI MI", split="test",  subjects=[1])

Download: OpenBMI

---

Preprocessing

All steps follow the fit / transform / fit_transform convention.
Each step accepts mne.BaseEpochs and returns mne.BaseEpochs (copy — never in-place).

from eeg_framework.preprocessing import *

pipe = Pipeline([
    ('CAR',        CAR()),
    ('RESAMPLE',   Resample(256.)),
    # ...
])

epochs_train_clean = pipe.fit_transform(epochs_train)
epochs_valid_clean = pipe.transform(epochs_valid)

---

Feature Extraction

All extractors return (X, y) where X is (n_epochs, n_features) and y is (n_epochs,).
The label mapping is captured during fit and reused during transform.

from eeg_framework.features import *

pipe_feat = FeaturePipeline([
    ('RAW', RawFeatures(flatten=False)),
    # ...
])
X_train, y_train = pipe_feat.fit_transform(epochs_train)
X_test,  y_test  = pipe_feat.transform(epochs_test)

---

Models

EEGNet

from eeg_framework.models import *

params = eegnet_params_from_epochs(epochs_train)
model  = build_eegnet(
    n_classes          = len(epochs_train.event_id),
    compile_model      = True,
    include_classifier = True,
    **params
)

model.fit(
    X_train, y_train,
    validation_split = 0.2,
    epochs           = 150,
)

---

Evaluation

from eeg_framework.evaluate import *

# Evaluation Metrics
metrics = compute_metrics(y_valid, y_pred, class_names=list(epochs_train.event_id))
# ...

# Evaluation Plots
# ...

---

Plotting

from eeg_framework.plotting import *

plot_erp_compare(epochs) 

# Othe Plots...
# ...

---

Dependencies

python 3.12.3
mne >= 1.0
numpy
scipy
scikit-learn
matplotlib
pandas
tensorflow >= 2.10   # for EEGNet