A fixed 1.5 factor reduced relative error estimate differences for five-pulse remote SEEG means at 20–500 ms in eight patients

We tested an error estimate for a mean SEEG response fitted from five early pulses. Public OpenNeuro ds004696 version 1.0.1 supplied eight patient sessions, 86 stimulation sites, and 886 remote bipolar readout pairs. Five separate later pulses tested each mean in the strict 20–500 ms window. Multiplying the early calibration residual by 1.5 reduced mean absolute relative difference from later error from 0.4244 to 0.3700. Each pair was divided by its first-five pulse energy. Patients and sites had equal weight. The mean gain was 0.05445, with a descriptive 95% patient bootstrap interval of 0.02635 to 0.08782. All eight patients improved. The exact two-sided patient sign-flip value was 0.0078125 under independent-patient and null sign-symmetry assumptions. In square microvolts, mean absolute difference increased from 2315.5 to 2550.0. Five patients became worse on this scale. The same forecast was used for both estimates. A proof gives an unbiased unnormalized estimate under a stable mean and independent equal-energy pulse noise. Simulations show failure with shared early noise or a later waveform change. Those assumptions were not established for the SEEG pairs. Source quality labels are retrospective and the later cohort is reused. The result supports a limited error-reporting method. It does not establish live BCI performance or a stimulation benefit.

stereoelectroencephalography · SEEG · sample mean · error estimation · calibration · brain-computer interface · pulse response · measurement error