Two causal multichannel gates changed held-out ripple-trigger F1 by less than 0.01 in five SEEG patients

We tested two causal multichannel gates for an 80–250 Hz ripple trigger using 120 seconds of interictal NREM SEEG from each of five patients in OpenNeuro ds003498. A same-shaft neighbor gate required an adjacent pair to exceed a threshold in the prior 20 ms; an other-shaft veto stopped triggers when the median envelope on other shafts exceeded a threshold. Thresholds were fitted on one 60-second half and evaluated on the other, in two folds, using author ripple marks as reference. Across 48 bipolar pairs, mean held-out F1 was 0.420 for the base trigger, 0.417 with the neighbor gate, and 0.421 with the veto. Patient-level mean F1 changes were −0.003 (95% CI −0.008 to 0.002) and +0.001 (95% CI −0.003 to 0.006), respectively. Both gates changed F1 by less than 0.01 in each patient. The neighbor gate reduced recall in all five patients and did not increase precision; the veto removed a mean of 0.4 triggers per pair per 60 seconds. Only five of 20 patients were analyzed. With this sample, the result bounds the observed effect size and is not a powered test of no effect. The original code, results, figure, and data provenance are included.

SEEG · ripple detection · causal trigger · closed-loop stimulation · multichannel gate · NREM sleep · OpenNeuro ds003498