Adding SEEG power through 200 ms increased squared log response time error by a mean 0.69% on later trials in eight recordings

We tested whether SEEG power through 200 ms improves a forecast of a later key response time. We used 2,026 trials from eight public recordings in OpenNeuro ds004473, version 1.0.2. A reference model used known task settings and past responses. Each person's first two thirds of trials trained a fixed ridge model. The remaining 678 trials tested it. A neural correction used beta and high gamma power from eight contact pairs selected in training. Its mean relative reduction in squared log response time error was -0.69%. Two of eight recordings improved. The 95% person bootstrap interval was [-1.89%, +0.74%]. A matched pre-cue control and fixed 100 and 250 ms variants gave no clear mean benefit. The result applies to these models and recording splits. It does not show that early SEEG lacks useful information. The package provides a reference test for future BCI timing methods. It contains features, source pins and hashes, fitted weights, forecasts, figures, and reproduction code. Fresh retrieval matched every used raw range and feature. Independent Fourier and ridge checks passed. This exploratory clinical cohort has no new-person, independent-day, or stimulation test.

stereoelectroencephalography · response time prediction · brain-computer interfaces · ridge regression · time series · OpenNeuro