One-pulse 12-way SEEG site identification retained accuracy at 64 Hz, 16 µV steps, and 50 ms in an eight-patient offline simulation

A bidirectional brain-computer interface must confirm which site it stimulated. We evaluated one-pulse identification of 12 stimulated stereoelectroencephalography (SEEG) pairs from same-shaft bipolar channels whose contacts were at least 5 mm from every stimulated contact. The analysis used 960 clinical single pulses from eight patients in OpenNeuro ds005448 v1.0.0 (12 sites and 10 pulses per site) and a 12-way leave-one-pulse-out nearest-centroid classifier. We simulated ADC rounding, blanking from -2 to +10 ms, causal boxcar decimation, and the decision window, changing one factor at a time. Reference median accuracy at 2048 Hz, floating-point precision, and a 15–145 ms window was 0.704. At 64 Hz, median accuracy was 0.688 (median ratio 0.990; lowest patient ratio 0.944; exact sign-flip p = 1.00); at 128 Hz it was 0.717 (ratio 1.023; lowest ratio 0.944; p = 0.34). At 32 Hz it decreased to 0.646 (ratio 0.936; lowest ratio 0.803; p = 0.031). A 16 µV ADC step retained median accuracy of 0.704 (lowest ratio 0.975; p = 1.00); 64 µV decreased median accuracy to 0.637 (ratio 0.928; lowest ratio 0.774; p = 0.031), and 256 µV to 0.500 (ratio 0.717; lowest ratio 0.375; p = 0.008). A 15–50 ms window retained accuracy (median 0.708; ratio 1.022; lowest ratio 0.975; p = 0.063). At 15–25 ms, median accuracy was 0.650 (ratio 0.976; lowest ratio 0.797; p = 0.51). Thus, in this offline simulation, 64–128 Hz sampling, ADC steps of 16 µV or less, and a 50 ms decision window retained performance relative to the 2048 Hz reference, while coarse ADC steps were the main tested risk. The recorded signals were already filtered by the clinical amplifier; blanking was assumed, channels 5–10 mm away could contain some direct-field response, only eight patients and 10 pulses per site were studied, and combinations of hardware limits were not tested. The planned 10 mm sensitivity check was not run.

SEEG · brain-computer interface · single-pulse stimulation · site identification · ADC resolution · sampling rate · intracranial EEG