Neurophysiologically meaningful motor imagery EEG simulated data

Description

This dataset contains different sessions of artificially generated motor imagery electroencephalographic (MI-EEG)-like signals. The data was generated based on PySimMIBCI [1], a framework for generating realistic MI-EEG signals by integrating neurophysiologically meaningful activity into biophysical forward models.

Specifically, MI was modeled by simulating an Event-Related Desynchronization (ERD) in the α band (9-11 Hz) over the contralateral motor area, representing the well-known brain dynamic during hand MI tasks. Right hand MI was differentiated from rest activity by a reduction in amplitude (ERD%) on the left hemisphere. For rest trials, idle α activity was simulated without ERD. The simulation includes three types of noise and artifacts: background noise, blink artifacts, and eye movement artifacts. For more information about the simulation process, please refer to [1].

Different sessions were simulated, with each session representing different simulated scenarios:

Self-Regulation Capabilities: To evaluate MI-BCI performance under different self-regulation capabilities, nine sessions were simulated with varying levels of ERD%, ranging from 50% to 10% in steps of 5%. These sessions reflect different user abilities to modulate α-band activity during MI, with higher ERD% indicating stronger MI modulation. The name of each session indicates the ERD% level, which was constant for all the trials of the session. For example, session S_30 represents an ERD% of 30%. It is important to mention that in all the simulated sessions, the subject performed the task correctly in every trial, meaning no failed trials were included.

Varying MI Modulation: An MI vs. Rest session was generated where MI trials were simulated with an ERD% sampled from a uniform distribution from 10% to 50%. This session simulates trials with varying levels of MI modulation within the same BCI session, providing a more continuous representation of user performance fluctuations during MI-BCI tasks. The name of this session is S_mix. A list of the ERD% corresponding to each trial is included at the S_mix_ERD_list.npy file.

Fatigue-Related Artifacts: An additional MI vs. Rest session that shows the impact of non-MI-related cortical activity, specifically fatigue-related effects. Fatigue was simulated by increasing frontal θ and parietal α power across the course of the session, as described in [1]. The first half of the trials were fatigue-free, while the second half gradually incorporated increasing fatigue levels. The MI trials in this session were simulated with a fixed ERD% of 50% ensuring consistent modulation ability throughout the session. The name of this session is S_fatigue.

Each session contains 100 trials per class (MI vs. Rest), of 4s duration each, simulated at 41 electrodes following the 10-5 electrode system, with a sampling frequency of 1000 Hz. It is important to mention that in all simulated sessions, the task wask correctly performed in the 100% of the trials, meaning no failed trials were included.

File Format:

Epoched data are provided as .fif files. This data can be easily read by mne.read_epochs function from the MNE-Python library.

Potential Use Cases:

This dataset is suitable for testing hypothesis during MI-BCI algorithms development, specially to study the impact of subject self-regulation capabilities and non-MI-related cortical activity on BCI efficacy.

References:

[1] C. M. Galván, R. D. Spies, D. H. Milone and V. Peterson, "Neurophysiologically Meaningful Motor Imagery EEG Simulation With Applications to Data Augmentation," in IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 32, pp. 2346-2355, 2024, doi: 10.1109/TNSRE.2024.3417311.