EEG Pipeline

An EEG pipeline for encoding and decoding analysis built using MNE python. The pipeline design is highly influenced by sklearn.pipeline.Pipeline and processing steps are passed as a list.

In this repository, we use the pipeline to perform cleaning, ICA, ERP peak calculation, Mass Univariate analysis and Decoding Analysis on subjects from ERPcore [1] for the P300 task.

Getting started

Quick Example

bids_root = os.path.join('data', 'P3')
bids_path = BIDSPath(subject='001',
                            session='P3',
                            task='P3',
                            datatype='eeg',
                            suffix='eeg',
                            root=bids_root)
pipeline = Pipeline(bids_path, verbose=logging.ERROR)
pipeline.load_data()
pipeline.make_pipeline([
        SimpleMNEFilter(0.1, 50, 'firwin'),
        CleaningData(self.bids_path),
        PrecomputedICA(self.bids_path)
])

Setup

Use the environment.yml file to create a virtual env and install mne-bids

Features

• Provides all the tools from MNE python.
• Use Multiprocessing for processing multiple subjects.
• Takes list of processing steps and applies them sequentially.

Classes

Note: The python files below provides basic structure. Once, I get some free time, I will refactor it. I would welcome and appritiate any pull requests.

• preprocessing.py: It contains classes for preprocessing steps. You can create your own preprocessing class by extending the base class.
• erpanalysis.py: It contains functions for performing ERP analysis.
• encoding_analysis.py: It contains functions for performing encoding analysis.
• decoding_analysis.py: It contains functions for performing decoding analysis.
• pipeline.py: It is a class responsible for performing analysis steps or orchestration. It also has some other helper functions and classes.
• test_pipeline.py: It is a test class for pipeline.py and can be used for testing processing steps.

Notebooks

Notebooks below have been moved to repository: eeg-analysis-notebooks

• preprocessing.ipynb: In this notebook, we will discuss about how to perform preprocessing steps like applying filters, removing bad channels, annotating bad segments from EEG data. We also discuss about performing ICA on EEG data for picking up and removing bad ICA components.

• erpanalysis.ipynb: In this notebook, we will disscuss peak to peak amplitude, difference between conditions and stimulus.

• encoding.ipynb: In this notebook, we will discuss and analyse how to fit a mass univariate model on EEG data.

• decoding.ipynb: In this notebook, we will disscuss and analyse classification scores based on Sensor-space decoding: decoding over time and decoding in sensor space data. This is highly influenced by MNE tutorials

Structure

+-- cleaning
|   +-- badChannels
        +-- ..files containing bad channels for 3-subjects
|   +-- badSegments
        +-- ..files containing bad segments for 3-subjects
+-- data
|   +-- sub_1
|   +-- ..
+-- docs
|   +-- ..
+-- images
|   +-- ...
|   +-- plots of discussions
+-- src

TODOs

• Major refactoring - change variable names, organize code. See the issue#1
• Replace list comprehension with generators
• Add Decoder based on Deep learning
• Performance testing / abalation studies

References

[1] Emily S. Kappenman, Jaclyn L. Farrens, Wendy Zhang, Andrew X. Stewart, Steven J. Luck, ERP CORE: An open resource for human event-related potential research, NeuroImage, Volume 225, 2021, 117465, ISSN 1053-8119

[2] EEG Preprocessing and analysis course, https://github.com/s-ccs/course_eeg_SS2021