Five-pulse weights with 50 percent covariance shrinkage increased later 2 to 30 Hz SEEG log power variance by 39.0 percent at 100 to 500 ms in eight STReEF patients

We tested whether nonnegative weights from five early stimulation pulses reduced later scatter in a SEEG log power monitor. The source was public STReEF, OpenNeuro ds005448 version 1.0.0. The sample had eight epilepsy patients, 96 conditions, and 960 pulses. We used nominal 2 to 30 Hz power at 100 to 500 ms after each pulse. Early shaft means and standard deviations were fixed before testing. The main weights used 50 percent correlation shrinkage toward identity and summed to one. Compared with equal weights, the mean patient later variance ratio was 1.390. This was a 39.0 percent increase, with a descriptive 95 percent whole-patient bootstrap interval of 15.0 to 64.6 percent. Six of eight patients had greater later variance. Calibration variance was 41.5 percent lower. Stronger fixed shrinkage reduced the later loss. Time and stimulation contact order changed together. The weights also changed the measured after-minus-before shift. The result measures output scatter. Neural estimation error and BCI benefit remain untested. This is an exploratory extension on an existing public cohort. All 960 raw ranges, 1,088,480 powers, 147 metadata files, 864 fitted models, and the result replay passed verification. The package preserves code, powers, source pins, fitted models, all checks, four figures, and reproduction instructions.

stereoelectroencephalography · brain-computer interfaces · covariance shrinkage · signal variance · electrical stimulation · power monitoring · reproducibility