A beta SEEG model with half the channel correlation gave an uncertain mean score change after stimulation in 20 patients

We tested a fixed Gaussian covariance model on public SEEG from OpenNeuro ds006519 version 1.0.0. The model kept half of the beta channel correlation measured during two seconds before stimulation. We compared it with a diagonal model on the last two seconds of the labeled post window. The test used 34 records from 20 patients. Mean gain was -0.0096 natural log units per channel and sample. Its 95% patient-bootstrap interval was from -0.0570 to +0.0220. Positive gain means a lower model score. Twelve patients had positive gains. A large loss in one patient changed the mean. A planned secondary model kept 10% of correlation. It gave positive beta gains in all 20 patients. Its mean gain was +0.0149. This secondary result is descriptive. All raw source files passed hash checks. All 374 rebuilt input arrays matched the archived inputs exactly. Independent raw voltage scores, patient inference, and result replay passed checks. The package preserves source code, fitted models, complete scores, figures, metadata, and retrieval instructions. This test measures a working Gaussian voltage model. It does not measure command accuracy or a causal stimulation effect.

stereo-electroencephalography · brain-computer interface · covariance shrinkage · Gaussian score · patient bootstrap · negative motor response