EEG-Dash

To leverage recent and ongoing advancements in large-scale computational methods and to ensure the preservation of scientific data generated from publicly funded research, the EEG-DaSh data archive will create a data-sharing resource for MEEG (EEG, MEG) data contributed by collaborators for machine learning (ML) and deep learning (DL) applications.

Data source

The data in EEG-DaSh originates from a collaboration involving 25 laboratories, encompassing 27,053 participants. This extensive collection includes MEEG data, which is a combination of EEG and MEG signals. The data is sourced from various studies conducted by these labs, involving both healthy subjects and clinical populations with conditions such as ADHD, depression, schizophrenia, dementia, autism, and psychosis. Additionally, data spans different mental states like sleep, meditation, and cognitive tasks. In addition, EEG-DaSh will incorporate a subset of the data converted from NEMAR, which includes 330 MEEG BIDS-formatted datasets, further expanding the archive with well-curated, standardized neuroelectromagnetic data.

Available data

The following datasets are currently available on EEGDash.

DatasetID	Participants	Files	Sessions	Population	Channels	Is 10-20?	Modality	Size
ds002181	20	949	1	Healthy	63	10-20	Visual	0.163 GB
ds002578	2	22	1	Healthy	256	10-20	Visual	0.001 TB
ds002680	14	4977	2	Healthy	0	10-20	Visual	0.01 TB
ds002691	20	146	1	Healthy	32	other	Visual	0.001 TB
ds002718	18	582	1	Healthy	70	other	Visual	0.005 TB
ds003061	13	282	1	Not specified	64	10-20	Auditory	0.002 TB
ds003690	75	2630	1	Healthy	61	10-20	Auditory	0.023 TB
ds003805	1	10	1	Healthy	19	10-20	Multisensory	0 TB
ds003838	65	947	1	Healthy	63	10-20	Auditory	100.2 GB
ds004010	24	102	1	Healthy	64	other	Multisensory	0.025 TB
ds004040	13	160	2	Healthy	64	10-20	Auditory	0.012 TB
ds004350	24	960	2	Healthy	64	other	Visual	0.023 TB
ds004362	109	9162	1	Healthy	64	10-20	Visual	0.008 TB
ds004504	88	269	1	Dementia	19	10-20	Resting State	2.6 GB
ds004554	16	101	1	Healthy	99	10-20	Visual	0.009 TB
ds004635	48	292	1	Healthy	129	other	Multisensory	26.1 GB
ds004657	24	838	6	Not specified	64	10-20	Motor	43.1 GB
ds004660	21	299	1	Healthy	32	10-20	Multisensory	7.2 GB
ds004661	17	90	1	Not specified	64	10-20	Multisensory	1.4 GB
ds004745	52	762	1	Healthy	64	?	Auditory	0 TB
ds004785	17	74	1	Healthy	32	?	Motor	0 TB
ds004841	20	1034	2	Not specified	64	10-20	Multisensory	7.3 GB
ds004842	14	719	2	Not specified	64	?	Multisensory	5.2 GB
ds004843	14	649	1	Not specified	64	?	Visual	7.7 GB
ds004844	17	481	4	Not specified	64	?	Multisensory	22.3 GB
ds004849	17	481	4	Not specified	64	?	Multisensory	0.077 GB
ds004850	17	481	4	Not specified	64	?	Multisensory	0.077 GB
ds004851	17	481	4	Not specified	64	?	Multisensory	0.077 GB
ds004852	17	481	4	Not specified	64	?	Multisensory	0.077 GB
ds004853	17	481	4	Not specified	64	?	Multisensory	0.077 GB
ds004854	17	481	4	Not specified	64	?	Multisensory	0.077 GB
ds004855	17	481	4	Not specified	64	?	Multisensory	0.077 GB
ds005034	25	406	2	Healthy	129	?	Visual	61.4 GB
ds005079	1	210	12	Healthy	64	?	Multisensory	1.7 GB
ds005342	32	134	1	Healthy	17	?	Visual	2 GB
ds005410	81	492	1	Healthy	63	?	?	19.8 GB
ds005505	136	5393	1	Healthy	129	other	Visual	103 GB
ds005506	150	5645	1	Healthy	129	other	Visual	112 GB
ds005507	184	7273	1	Healthy	129	other	Visual	140 GB
ds005508	324	13393	1	Healthy	129	other	Visual	230 GB
ds005509	330	19980	1	Healthy	129	other	Visual	224 GB
ds005510	135	4933	1	Healthy	129	other	Visual	91 GB
ds005511	381	18604	1	Healthy	129	other	Visual	245 GB
ds005512	257	9305	1	Healthy	129	other	Visual	157 GB
ds005514	295	11565	1	Healthy	129	other	Visual	185 GB
ds005672	3	18	1	Healthy	64	10-20	Visual	4.2 GB
ds005697	52	210	1	Healthy	64	10-20	Visual	67 GB
ds005787	30	?	4	Healthy	64	10-20	Visual	185 GB

Data format

EEGDash queries return a Pytorch Dataset formatted to facilitate machine learning (ML) and deep learning (DL) applications. PyTorch Datasets are the best format for EEGDash queries because they provide an efficient, scalable, and flexible structure for machine learning (ML) and deep learning (DL) applications. They allow seamless integration with PyTorch’s DataLoader, enabling efficient batching, shuffling, and parallel data loading, which is essential for training deep learning models on large EEG datasets.

Data preprocessing

EEGDash datasets are processed using the popular BrainDecode library. In fact, EEGDash datasets are BrainDecode datasets, which are themselves PyTorch datasets. This means that any preprocessing possible on BrainDecode datasets is also possible on EEGDash datasets. Refer to BrainDecode tutorials for guidance on preprocessing EEG data.

EEG-Dash usage

Install

Use your preferred Python environment manager with Python > 3.9 to install the package.

• To install the eegdash package, use the following temporary command (a direct pip install eegdash option will be available soon): pip install -i https://test.pypi.org/simple/eegdash
• To verify the installation, start a Python session and type: from eegdash import EEGDash

Data access

To use the data from a single subject, enter:

from eegdash import EEGDashDataset
ds_NDARDB033FW5 = EEGDashDataset({'dataset': 'ds005514', 'task': 'RestingState', 'subject': 'NDARDB033FW5'}, description_fields=['sex'])

This will search and download the metadata for the task RestingState for subject NDARDB033FW5 in BIDS dataset ds005514. The actual data will not be downloaded at this stage. Following standard practice, data is only downloaded once it is processed. The ds_NDARDB033FW5 object is a fully functional BrainDecode dataset, which is itself a PyTorch dataset. This tutorial shows how to preprocess the EEG data, extracting portions of the data containing eyes-open and eyes-closed segments, then perform eyes-open vs. eyes-closed classification using a (shallow) deep-learning model.

To use the data from multiple subjects, enter:

from eegdash import EEGDashDataset
ds_ds005505rest = EEGDashDataset({'dataset': 'ds005505', 'task': 'RestingState'}, target_name='sex')

This will search and download the metadata for the task 'RestingState' for all subjects in BIDS dataset 'ds005505' (a total of 136). As above, the actual data will not be downloaded at this stage so this command is quick to execute. Also, the target class for each subject is assigned using the target_name parameter. This means that this object is ready to be directly fed to a deep learning model, although the tutorial script performs minimal processing on it, prior to training a deep-learning model. Because 14 gigabytes of data are downloaded, this tutorial takes about 10 minutes to execute.

Automatic caching

EEGDash automatically caches the downloaded data in the .eegdash_cache folder of the current directory from which the script is called. This means that if you run the tutorial scripts, the data will only be downloaded the first time the script is executed.

Education -- Coming soon...

We organize workshops and educational events to foster cross-cultural education and student training, offering both online and in-person opportunities in collaboration with US and Israeli partners. Events for 2025 will be announced via the EEGLABNEWS mailing list. Be sure to subscribe.

About EEG-DaSh

EEG-DaSh is a collaborative initiative between the United States and Israel, supported by the National Science Foundation (NSF). The partnership brings together experts from the Swartz Center for Computational Neuroscience (SCCN) at the University of California San Diego (UCSD) and Ben-Gurion University (BGU) in Israel.