Abstract
The Pan2025 dataset contains electroencephalography (EEG) signals from 10 subjects performing motor imagery (MI) tasks across two experimental sessions. This dataset facilitates research on cross-session variability in MI-EEG signals and supports development of robust brain-computer interface (BCI) systems.
Dataset Composition
The dataset includes EEG recordings from 10 subjects across two sessions, featuring left-handed and right-handed MI tasks with visual cues. Data was acquired using:
- Neuroscan SynAmps2 amplifier
- 28 scalp electrodes (10-20 system)
- 250Hz sampling rate (after downsampling)
- Band-pass filtering (0.01-200Hz)
Data is stored in MATLAB format with subject/session labeling.
Participants
The cohort consists of 10 individuals (3 females, 7 males) aged 22-25 years, including 2 left-handed participants. All subjects were neurologically healthy with no movement disorders.
Experimental Paradigm
Referenced from: Pan et al. (2023). Riemannian geometric and ensemble learning for decoding cross-session motor imagery electroencephalography signals. Journal of Neural Engineering [DOI]
Each session (~180 trials) contained three phases:
- Rest Period (4s): Relaxation without mental tasks
- Task Period (4s): MI execution (left/right hand)
Conducted in a controlled environment. For Session 2:
- First 30 trials: Training
- Remaining trials: Testing with online feedback
Note: Trial counts vary across sessions and subjects.
Data Acquisition and Preprocessing
Technical specifications:
- Original sampling rate: 1000Hz
- Band-pass filter: 0.01-200Hz
- Notch filter: 50Hz (powerline noise removal)
- Final sampling rate: 250Hz (after downsampling)
Data Structure
MATLAB struct containing:
data
: 3D matrix [n_channels × n_samples × n_trials]
label
: Trial labels [n_trials] (1: left-hand, 2: right-hand)
fs
: Sampling frequency (250Hz)
period
: Trial duration (seconds)
chaninfo
: Channel metadata (cell array)
(2025-07-05)