Early RMS did not lower mean late SEEG forecast error in fixed linear models with five calibration pulses per site in eight patient sessions

We tested a fast voltage forecast for a brain-computer interface that writes to the brain. The source was public SEEG data from OpenNeuro ds004696 version 1.0.1. We used eight patient sessions, 86 stimulation sites, and 886 bipolar readout pairs. The first five good pulses at each site set a fixed model. The final five tested it. Models used only calibration sets that were complete before the current site model was fixed. The target was centered voltage RMS at 100–500 ms. Adding RMS at 20–100 ms to pre-pulse RMS increased mean squared log-RMS error by 1.55%. The 95% patient bootstrap interval for error reduction was -6.50% to 1.67%. Error was lower in 5 of 8 patients. This test did not establish a mean benefit from the early feature. The matched pre-pulse check gave a descriptive 4.30% mean error reduction. The package supplies matched background data, exact trial products, all forecasts, source checks, code, and figures. The result applies to these fixed models and clinical sessions. It does not establish the effect of stimulation or test an online controller.

stereoelectroencephalography · SEEG · voltage forecast · brain-computer interface · RMS · calibration · patient bootstrap