Ocular Blink Artifact Removal via Picard ICA & ICLabel
Eye blinks and rapid saccades produce massive fronto-polar voltage deflections (up to 150 µV on $F_{p1}, F_{p2}$) that volume-conduct across the entire scalp, corrupting frontal theta and sensorimotor beta power.
Why Picard Fast Non-Linear ICA Outperforms Infomax
Standard Extended Infomax ICA takes up to 20 minutes on high-density 64-channel recordings and frequently converges on local sub-optimal non-linear extrema. The Meridian Engine employs Picard (Preconditioned Independent Component Analysis for Real Data), delivering:
- 10x Faster Convergence: Processes full 64-channel clinical recordings in seconds.
- Guaranteed Non-Linear Separation: Cleaner separation of frontal ocular dipoles from frontal-midline theta ($F_z$) oscillations.
- Automated ICLabel Scoring: Each independent component is scored against a pretrained deep neural network evaluating topographies and power spectral densities. Components with Eye probability $P \ge 0.90$ are zeroed.
Verified Retention of Neural Signals
In validation tests on clinical EEG data, zeroing blink ICs eliminated 37.9% of total signal variance (the exact ocular contamination) while retaining 99.4% of underlying occipital and sensorimotor alpha power ($8–12\text{ Hz}$).
Eliminate Ocular Contamination from Your Data
Stop discarding blink-contaminated epochs. Meridian cleans your files while preserving every valuable millisecond of neurological recording.
Submit Dataset for Picard ICA Cleaning →