A fixed voltage RMS model gave a mean late SEEG prediction gain of 0.08% on separate pulse halves in five patients

We tested whether voltage size before a pulse improved prediction of late voltage after the pulse. We used public SEEG data from five patients with epilepsy in OpenNeuro ds004457 version 1.0.2. RMS means root mean square. The feature was log RMS of locally centred bin means from 500 to 300 ms before the pulse. The outcome used 100 to 300 ms after the pulse. Each stimulation block and receiving contact supplied ten selected pulses. We trained on the first five and tested on the last five. The fixed model added one common slope per patient to each link's training mean. Across 436 links, the equal-patient mean test error reduction was 0.083%. Four patients had worse predictions; one had better predictions. The descriptive patient resample range was -2.19% to 3.46%. The exact patient sign-flip p value was 1.00. These results give no clear evidence of mean prediction benefit for this feature and model. They do not establish equivalence. This retrospective pulse split does not test forward prediction through the session. The package has input bins, predictions, code, source hashes, seven planned checks, and instructions. A fresh raw-data rebuild matched every input array. The result applies to this model and these five patients.

SEEG · intracranial EEG · electrical stimulation · voltage RMS · prediction · pulse response · epilepsy · patient-level analysis