Replication Data for: A cross-session motor imagery EEG dataset

# Pan2023 Dataset Documentation # This is a replication of the "A cross-session motor imagery EEG dataset" dataset, the .mat file version is v7.0. ## Abstract The Pan2023 dataset is a collection of electroencephalography (EEG) signals from 14 subjects performing motor imagery (MI) tasks across two sessions. The dataset aims to facilitate the study of cross-session variability in MI-EEG signals and to support the development of robust brain-computer interface (BCI) systems. ## Dataset Composition The dataset encompasses EEG recordings from 14 subjects, each participating in two sessions. The sessions involve MI tasks with visual cues for left-handed and right-handed movements. Data acquisition was performed using a Neuroscan SynAmps2 amplifier, equipped with 28 scalp electrodes following the international 10-20 system. The EEG signals were sampled at a frequency of 250Hz, with a band-pass filter applied from 0.01 to 200Hz to mitigate power line noise. The collected data is stored in Matlab format, labeled by subject and session number. ## Participants The participant cohort includes 14 individuals (five females), aged 22 to 25, with two reporting left-handedness. All subjects were screened for neurological and movement disorders, ensuring a healthy participant profile for the study. ## Experimental Paradigm Each experimental session comprised 120 trials, segmented into three distinct phases: Rest, Preparation, and Task. During the Rest Period (2 seconds), subjects were instructed to remain relaxed without engaging in mental tasks. The Preparation Period (1 second) involved a 'Ready' cue on the monitor, prompting subjects to focus and prepare for the upcoming MI task. The Task Period (4 seconds) required subjects to perform the MI task, visualizing the movement corresponding to the provided cues, either left or right-handed. This paradigm was designed to occur in a controlled, distraction-free environment. ## Data Acquisition and Preprocessing EEG signals were captured using a Neuroscan SynAmps2 amplifier and 28 scalp electrodes positioned per the 10-20 system. The sampling rate was set at 1000Hz, and a band-pass filter from 0.01 to 200Hz and a notch filter at 50Hz were employed to exclude power line interference. The signals were downsampled to 250Hz and archived in Matlab format, systematically named by subject and session identifiers. ## Data Structure The dataset's structure is encapsulated in a Matlab file, comprising a struct with the following components: - `data`: A 3D matrix (`[n_trials, n_channels, n_samples]`) containing the EEG signals. - `label`: A vector (`[n_trials]`) denoting each trial's label (1 for left-handed, 2 for right-handed movement). - `trial_info`: A struct detailing each trial's phase (1 for Rest, 2 for Preparation, 3 for Task), the visual cue (1 for left-handed, 2 for right-handed movement), and the subject's identifier.