A 20 percent trimmed mean raised expert label AUC from 0.892 to 0.923 for ten pulses and a 10 to 90 ms score with each method's baseline SD in UMCU25 SEEG data

We tested whether a robust pulse average improves the rank of expert response labels in public SEEG data. We used release 1.0.0 of OpenNeuro ds004774. The data contain 630 surface stimulation pulses and six depth readout contacts in each of two recordings. The fixed score uses the largest absolute voltage from 10 to 90 ms, divided by each method's baseline standard deviation. With ten pulses, a 20 percent trimmed mean raised label AUC from 0.892 to 0.923 in UMCU25. The paired site bootstrap change bounds were 0.008 to 0.059. Both UMCU21 expert groups had smaller changes and bounds that included zero. Control scores before the pulse also changed. Thus, the gain cannot be assigned only to better capture of an evoked neural response. These results are conditional on the recordings and expert labels. They do not establish a general brain interface gain or a lower pulse budget. The package contains native voltage arrays, labels, fixed code, all comparison tables, resampling draws, three figures, pinned raw retrieval instructions, and completed check receipts.

SEEG · electrical stimulation · evoked responses · trimmed mean · expert labels · area under the ROC curve · signal processing