What can you do?

The library allows you to load and preprocess EEG recordings and align them to a common template, i.e. a set of ordered channels. Each file of the selected datasets will have the same channels, even if data comes from different channel systems, with different numbers of electrodes. The preprocessed and template-aligned files can then be imported and used in Deep Learning frameworks such as PyTorch, thus BIDSAlign is thought to be used as an initial step in a deep-learning project involving EEG data.

Channel systems currently supported are:

• HydroCel Geodesic Sensor Net (128,129,256,257 channels).
• Standard 10-20, 10-10, 10-5 systems (up to 344 channels).

It comes with a dedicated GUI, please see the attached detailed documentation.

Preparation Steps

In order to successfully use BIDSAlign, check the following steps.

Datasets List

You have to create a folder where you will store all the datasets present in the DATASET_INFO.tsv file, which is structured in the following way:

dataset_number_reference	dataset_name	dataset_code	channel_location_filename	channel_system	channel_reference	channel_to_remove	eeg_file_extension	samp_rate	line_noise
1	HBN_EO_EC	ds004186	loaded	GSN129	CZ		.set	500
2	Test_Retest_Rest	ds004148		10_10	FCZ		.vhdr	500

Please remember that the name of the folder where the dataset is stored must corresponds to the value in column dataset_code.

The library already comes with all the information needed for processing 20 public datasets. Some of them are large and commonly used in deep learning such as the Healthy Brain Network HBN EO/EC, the Two Decades–Brainclinics Research Archive for Insights in Neurophysiology TDBRAIN and the Max Planck Institute LEMON dataset MPI-LEMON.

BIDS Format

This library can preprocess datasets structured with both BIDS and non-BIDS format, thus is expected in input a dataset structured as shown in BIDS-EEG. However you can use the function create_dataset_architecture.m to change in-place the folder structure of the dataset. As specified by the BIDS format, the participants.tsv file is recommended and should be stored inside the dataset folder.

Here an example of how the folder structure and the workflow can be organized.

Channel Location

Many times the channel location is absent or even wrong when downloading public data. BIDSAlign helps you, setting the column channel_location_filename, in the following way:

• If a specific channel location should be used for the entire dataset, please enter the filename.
• If the eeg files have already the channel coordinates inside the EEG structure, please enter 'loaded'.
• If the default channel location can be used for the dataset, please enter 'default' and it will use the default channel locations present in EEGLAB.
• If no channel location is specified and the field is left empty, than _electrodes.tsv files are searched and eventually used.

Usage Modalities

You can use the library in three ways:

1. Preprocess a specified file. You have to specify the dataset name, from which the file is taken, the filename and the corresponding filepath. The modality is activated with the variable single_file.

single_file = true;
dataset_name = ['UC_SD'];
raw_filename = ['sub-hc10_ses-hc_task-rest_eeg.bdf']; 
raw_filepath = ['/Users/.../Datasets/ds002778/sub-hc10/ses-hc/eeg/'];

2. Preprocess an entire dataset. You have to specify the dataset. In this case you can process the entire dataset, a portion, only some subjects or some sessions, or even preprocess specific groups and/or specific task.

single_file = false;
dataset_name = ['UC_SD'];

3. Preprocess all the datasets contained in DATASET_INFO.tsv You have to left empty the dataset name and set single file as false.

single_file = false;
dataset_name = [];

Preprocessing

The following preprocessing steps are currently available in BIDSAlign:

1. Channels removal.
2. Segment removal (first and last specified seconds).
3. Baseline removal.
4. Resampling.
5. Filtering.
6. Independent Component Analysis (ICA) and Automatic IC rejection with MARA or ICLabel.
7. Notch-Filtering.
8. Artifact Subspace Reconstruction (ASR) can be used independently in two ways: for removing bad channels and/or for removing/reconstructing bad time windows.
9. Interpolation of previously removed bad channels.
10. Rereference.

Please note that EEG data are assumed to be saved in $\mu V$.

Saving and Visualisation

BIDSAlign allows you to do a proper visualisation of the data and perform statistical analysis. In order to use the visualisation functions, please save the .set or the .mat folders by specifing save_info.set_label as group _ pipeline; for example group could be 'A' indicating Alzheimer's and pipeline could be 'ICA' indicating that you have performed Independent Component Analysis, thus A_ICA.

The library comes with three functions:

1. groups_visualization: you can compare more groups for a single pipeline, or viceversa; you can also specify the single filename to be visualised.
2. ERP_visualization: you can plot the average and grand-average ERP for a group of patients, for multiple events. If there is only one event, scalp topographies of channels activation in time are also shown.
3. template_comparison: you can see the differences between the topographies obtained from two channel location and the effects of the conversion file.

Compatibility

The library was written in MATLAB 2023b, EEGLAB 2023.0 and requires the following plug-in:

• "Biosig" v3.8.3
• "FastICA" v25
• "Fileio" v20240111
• "ICLabel" v1.4
• "MARA" v1.2
• "bva-io" v1.73
• "clean_rawdata" v2.91
• "dipfit" v5.3
• "firfilt" v2.7.1
• "eegstats" v1.2

BIDSAlign can automatically install all these plugins if not already included in your EEGLAB installation. (The dependecy list is checked when bidsaling or bidsalign nogui is called.) You can also run a fresh installation of EEGLAB and all the required plugins by running the install_eeglab_from_scratch( 'target_path' ) function. It is suggested to use as target_path the 'MATLAB' directory or one of its subdirectories.

Moreover interally it uses two libraries:

• For the non-parametric permutation t-test hfASDmodules: Reorganization of functionally connected subnetworks in autism.
• For the iaf calculation: restingIAF

If you want to avoid downloading these additional packages, please set paf=false and test_parametric=true when using the visualisation function groups_visualization.

Contributors:

• M.Sc. Andrea Zanola
• Eng. Federico Del Pup
• Prof. Camillo Porcaro
• Prof. Manfredo Atzori