Synthetic EEG with Known Motor Cortex Sources for Source Localization (19 channels)
Description
Description This dataset contains 20,000 synthetic 19-channel electroencephalogram (EEG) segments. For every segment the cortical sources that generated it are known exactly by construction (full ground truth). It was designed to train and objectively evaluate EEG source localization methods, especially deep-learning models, that aim to identify active regions of the primary motor cortex during motor tasks. Each example (individual) is a 2-second EEG segment sampled at 200 Hz (400 time samples), recorded from 19 electrodes of the international 10-20 system (Fp1, Fp2, F7, F3, Fz, F4, F8, T7, C3, Cz, C4, T8, P7, P3, Pz, P4, P8, O1, O2). The montage, sampling rate, 0.5–30 Hz band-pass filter and average reference replicate a real clinical acquisition device. Each example was generated by 1 to 4 simultaneously active cortical dipoles. The dipoles were chosen at random from 551 candidate locations on the precentral gyrus of both hemispheres (263 left, 288 right), used as an anatomical proxy of the primary motor cortex. Inputs (what a model receives) X (20000 × 19 × 400, float32, volts): scalp EEG. Axes are example × channel × time. Outputs / ground truth (what a model should recover) active_idx (20000 × 4, int32): indices of the active dipoles within the 8,196-point cortical source space. Only the first n_active entries are valid; unused entries are −1. n_active (20000, int32): number of active sources (1–4). S_active (20000 × 4 × 400, float32, A·m): time course of each active source, aligned with active_idx. dipole_pos (8196 × 3, float32, meters): (x, y, z) position of every dipole in head coordinates. Coordinates and depth of each active source can be looked up from active_idx. The label can be used at three levels of detail: the exact dipole (which of the 551 candidates), its 3-D coordinates (error measured in mm), or the cortical zone that contains it. Zones can be defined freely, for example by k-means clustering of the candidate positions, without regenerating the data. Sources are stored sparsely (only the active dipoles), so the full dataset fits in ~650 MB. Files: eeg19_clinical_v3.npz: the dataset (NumPy compressed archive). eeg19_clinical_v3.yaml: the configuration file used to generate it. README.md: loading instructions.
Files
Institutions
- Autonomous University of QueretaroQuerétaro, Querétaro City