This replays real 64-channel EEG from people imagining left- or right-hand movement, from the PhysioNet EEG Motor Movement/Imagery dataset. Each electrode's μ (8–12 Hz) and β (13–30 Hz) power is computed live by an FFT in your browser.
The rotating head is a topographic map: every point on the scalp is a GPU-interpolated blend of the 64 electrode values (hand-written WGSL). When someone imagines a movement, the opposite motor cortex desynchronises: its μ/β power drops, which the decoder reads from C3 vs C4 to call left vs right.
Honest notes: this is an offline replay of recorded data with a live read-out, a BCI-style demo, not a closed-loop BCI. The decoder is a physiology-based μ/β ERD contrast over the C3/C4 motor region (2 s window), not a trained classifier: measured ~60–66% on 2-class left/right (chance = 50%). The rotating head shows raw band power (per-recording scale); the decoder uses baseline-relative ERD. The actual cue is the true instruction, shown for comparison. Data is real recordings (imagined left/right fist), reformatted offline from EDF+, nothing synthesised.
…
Each map is the μ/β desynchronisation unique to that hand: its ERD minus the other's, which cancels activity common to both hands and leaves the lateralised motor component. The bright patch then sits on the side opposite the imagined hand (left hand → right cortex): the contralateral signal the decoder reads. Top-down, nose up.
…
CortexCast renders the scalp topomap with WebGPU. Your browser didn't expose it.
Try Chrome or Edge 113+, or Safari 18+.