Neural Noise

Multimodal EEG and fNIRS dataset for motor imagery (upper-limb joints)

2 min read

  • A multi-modal dataset of EEG and fNIRS recordings for motor imagery of multiple joint types in the unilateral upper limb was published in Nature (June 2025).1
  • The open dataset supports EEG-based BCI and fNIRS methods and multimodal decoding benchmarking.1 1

Weekly enrichment (2026-07-20)

  • The dataset (Yi, Chen, Wang et al., Scientific Data 12, 953, 2025) provides simultaneously recorded EEG and fNIRS from 18 subjects performing motor imagery of the right upper limb.23
  • It covers eight MI tasks spanning four joint types: hand open/close, wrist flexion/extension, wrist abduction/adduction, elbow pronation/supination, elbow flexion/extension, shoulder pronation/supination, shoulder abduction/adduction, and shoulder flexion/extension.23
  • Each subject contributed 320 trials (40 per MI task), yielding a total of 5,760 trials across the dataset.3
  • EEG was recorded with a 64-channel whole-head cap on a Neuroscan SynAmps2 amplifier at 1000 Hz, referenced to the left mastoid (M1), with electrode impedances kept below 10 kΩ.24
  • EEG acquisition applied a 50 Hz notch filter and a 0.5-100 Hz band-pass to cover the alpha and beta bands most relevant to motor imagery.24
  • fNIRS was collected with a NIRx NIRScout system using 8 sources and 8 detectors at a 7.8125 Hz sampling rate, arranged in an 8x8 grid over the left hemisphere per the international 10-5 system, producing 24 channels.24
  • Data validity was verified both through EEG/fNIRS activation patterns and through classification performance, supporting use for hybrid EEG-fNIRS decoding benchmarks.23
  • Raw recordings are openly hosted on figshare (DOI 10.6084/m9.figshare.24123303) with loading, plotting, and decoding code released in the FJMI_EEG_fNIRS GitHub repository (MATLAB/Python).5
  • The multi-joint, same-limb design targets more intuitive and higher-dimensional MI-BCI control than the common binary left/right paradigms, and complements newer directional hybrid datasets such as MIND (30 subjects, 64-channel EEG at 1000 Hz, 51-channel fNIRS at 47.62 Hz).26

Footnotes

  1. https://news.google.com/rss/articles/CBMiX0FVX3lxTE9Pc3Bfa3d0czlPTk01ZU00bzRaWnV3dXFBVmhzYm0tT0NhaWZoVWlXV0lRYTgzMVVWSU84Zm9Cb2FxR0JKSE9VcDh6Q2FkTGp2RkZsV1VwUm1VYWs5NjA0?oc=5 ↩ ↩2 ↩3

  2. https://www.nature.com/articles/s41597-025-05286-0 ↩ ↩2 ↩3 ↩4 ↩5 ↩6 ↩7

  3. https://doi.org/10.1038/s41597-025-05286-0 ↩ ↩2 ↩3 ↩4

  4. https://preview-www.nature.com/articles/s41597-025-05286-0 ↩ ↩2 ↩3

  5. https://github.com/Bill-Chen-BJ/FJMI_EEG_fNIRS ↩

  6. https://arxiv.org/abs/2602.04299 ↩

Graph View