The pulse-locked SEEG response falls within 2 s and plateaus, but the model does not improve held-out predictions

We measured the time course of the pulse-locked SEEG response during 43.2/50 Hz stimulation trains in 267 responsive bipolar channels from 22 dataset-subjects and 31 runs (35 input runs; 4 lacked a responsive channel or had a train shorter than 2 s). The data came from OpenNeuro ds005169 (15 subjects; visual-response sites) and ds006519 (7 subjects; negative-motor-response sites), using archived per-run features from prior SCI contributions and the same channel selection and noise correction. We fit an exponential-to-plateau model to noise-corrected polarity-sum response RMS in 0.5-s bins, normalized to the first bin. The group curve had tau 1.06 s (95% bootstrap CI 0.63–2.31) and plateau 0.67 (0.58–0.74). Median subject fits were tau 1.48 s (0.99–3.07) and plateau 0.55 (0.47–0.64); dataset-specific fits were similar. Most decrease occurred in the first 2 s, with response near 0.6–0.7 of its first value through the 5-s train. The polarity-difference artifact control stayed flat (median plateau 0.95, CI 0.88–1.11). For 263 channels with at least 6 bins, median tau was 1.33 s (IQR 0.42–5.0); 44% had tau <1 s and 22% had plateau <0.1. Tau and plateau did not correlate with distance to the stimulated pair (Spearman rho −0.01 and 0.03). In a held-out test of 22 subjects, the exponential-plateau model did not predict the last second better than a linear model (median absolute error 0.110 vs 0.147, p=0.11), an exponential-to-zero model (p=0.87), or carry-forward (error 0.095, p=0.39). Thus, the group curve has a plateau shape, but this model was not shown to outperform simple alternatives for an individual subject. Subject fits were poorly identified for 7 of 15 ds005169 subjects: 3 had tau at the 0.05-s lower bound and 4 had plateau near zero with long tau. Resolution was 0.5 s, trains lasted at most 5 s, and a slower decrease after 5 s remains possible. The pooled tests treat subjects in the two datasets as different people; IDs sub-13, sub-14 and sub-18 occur in both datasets, so independence is uncertain if these are the same people. Four subjects had only one responsive channel.

SEEG · intracranial stimulation · pulse-locked response · time course · evoked response · neural adaptation · stimulation train · exponential model