Neural Noise- HEFMI-ICH is a hybrid EEG-fNIRS motor imagery dataset for brain-computer interface research in intracerebral hemorrhage patients.1
- The dataset supports BCI methods development in a clinically relevant population with clear implementation path.1 1
Weekly enrichment (2026-07-20)
- HEFMI-ICH is described in Scientific Data (Nature Portfolio, 2025; PMID 41253791) by Shi, Chen, Zhao et al. as the first hybrid brain-computer interface dataset built specifically for intracerebral hemorrhage (ICH) rehabilitation research.2
- The dataset comprises synchronized recordings from 17 healthy subjects and 20 patients with ICH performing standardized left- and right-hand kinesthetic motor imagery.2
- Signals were captured with 32-channel EEG sampled at 256 Hz and 90-channel fNIRS sampled at 11 Hz, acquired simultaneously and time-aligned via E-Prime 3.0 event markers.2
- Hardware included a g.tec g.HIamp EEG amplifier (Graz, Austria) and a NirScan continuous-wave fNIRS system (Danyang Huichuang Medical Equipment, China), using a custom hybrid cap (Model M, 54–58 cm) with 32 EEG electrodes plus 32 optical sources and 30 photodetectors paired at 3 cm separations to form the 90 fNIRS channels.2
- The block-design paradigm ran at least two sessions per subject with 30 trials per session (15 left- and 15 right-hand motor-imagery trials), totaling 60 or more trials, preceded by baseline recording (1 min eyes-closed, 1 min eyes-open).2
- To improve motor-imagery vividness, a grip-strength calibration used a dynamometer and stress ball with repeated ~5 kg maximal efforts at roughly one contraction per second before acquisition.2
- The release provides three components — raw data (EEG in HDF5; fNIRS in NIRS/HCX), preprocessed trial-ready data, and clinical participant characteristics — with fNIRS light intensity converted to oxy-/deoxy-hemoglobin via the modified Beer-Lambert law and an ERD/ERS pipeline (0.5–30 Hz band-pass) implemented in Python’s MNE.2
- The authors also released a unified deep-learning classification framework using feature-level fusion with adaptive weighting across the two modalities, and the full dataset is publicly hosted on Figshare for algorithm validation and clinical BCI development.23