Initial response strength did not predict response decrease during 43–50 Hz SEEG stimulation trains

During 43.2–50 Hz stereo-electroencephalography (SEEG) stimulation trains, pulse-locked responses decrease over 5 s. This analysis tested whether strongly driven channels lose a larger fraction of their response than weakly driven channels, which would flatten the spatial activation pattern, or whether channels show a similar fractional decrease (a pure gain change). It used archived per-run features from 10 subjects and 134 channels in two OpenNeuro datasets: ds005169 (visual-percept sites, 50 Hz) and ds006519 (negative-motor-response sites, 43.2 Hz). Strength was log SNR in the first 0.5 s; decrease was log noise-corrected power in the last 1 s divided by power in the 0.5–1.0 s bin. These measures used separate bins. The primary statistic was the mean of per-subject Spearman correlations, compared with a 500-draw parametric pure-gain null. Median power ratio was 0.55 (amplitude ratio 0.74). The observed group correlation was −0.042, compared with null mean 0.001 (95% range −0.206 to 0.213; one-sided p = 0.34). Six of 10 subjects were below their null mean (sign-flip p = 0.33). The instrumental-variable slope was 0.96 (95% confidence interval for the median 0.83–1.08; Wilcoxon test against 1, p = 0.63), consistent with a pure gain change. Distance-adjusted correlation was 0.08 (p = 0.63). In a sensitivity analysis using SNR > 3, the correlation was −0.10 (12 subjects, 235 channels; null 95% range −0.14 to 0.16; p = 0.11), and the median instrumental-variable slope was 0.95 (p = 0.42). In an exploratory ds005169-only analysis, correlation was −0.24 (4 subjects; p = 0.04); with SNR > 3 it was −0.27 (6 subjects; p = 0.02). The ds006519-only correlation was +0.09 (6 subjects; p = 0.74). The ds005169 result was not planned; after correction for two datasets (alpha 0.025), it was not significant in the primary SNR > 5 analysis and was significant in the SNR > 3 analysis. The primary analysis agrees with a pure gain change but cannot exclude small compression. The group null range is about ±0.21, and the exploratory visual-site result needs a test in more subjects. Limits include 10 subjects, 6–31 channels per run, a parametric noise model, one train per run, responses measured 5–18 ms after each pulse, bipolar montage, different targets and frequencies, and strength measured as SNR rather than absolute amplitude.

stereo-electroencephalography · intracranial stimulation · response adaptation · signal-to-noise ratio · pure gain · compression · neural responses