Background forecasts reduced mean scaled error for four SEEG voltage bins at 20–100 ms in eight selected patients with mean removal

Voltage before a stimulation pulse improved prediction of later recorded SEEG voltage in the main fixed test. This fixed test used public OpenNeuro ds005448 data from eight selected patients with epilepsy. A background model used only activity before pulses at 22 earlier sites. The test used 96 later pulses at 24 other sites. The target contained four 20 ms voltage means from 20 to 100 ms, with their mean removed. The main comparison added the current background forecast to the same fixed template. The mean patient reduction in scaled error was 1.247 percent. The 95 percent patient bootstrap interval was 0.687 to 1.920 percent. Seven of eight patients had a positive gain. The two-sided sign-flip value was 0.015625, under an independent, sign-symmetric patient null model. A before-pulse control also had a positive mean gain. Later windows and raw voltage error left the gain uncertain. This result concerns recorded voltage, rather than an identified neural source. The patients were reused from prior studies. The package preserves compact inputs, raw examples, fixed source versions, code, scores, and reproduction instructions.

stereo-electroencephalography · SEEG · voltage prediction · background activity · electrical stimulation · prediction error · epilepsy · OpenNeuro ds005448