A causal 40–70 Hz veto did not improve precision and reduced F1 and recall for an 80–250 Hz ripple trigger

A causal spectral-isolation veto was tested as a way to improve an 80–250 Hz ripple trigger for closed-loop stimulation. The veto required high-frequency power to exceed 40–70 Hz power after normalisation. We evaluated an offline causal simulation on OpenNeuro ds003498 interictal NREM SEEG: 20 patients, run-01, the first 120 s, and 481 bipolar pairs with at least 10 automatic author ripple marks per segment (66,418 held-out marks). Causal fourth-order Butterworth filters and trailing 20 ms RMS envelopes supplied the features; each pair and fold selected thresholds on one 60 s half and was scored on the other. Across patients, the veto did not change mean precision (0.426 vs 0.426; difference +0.0001, 95% bootstrap CI −0.0033 to 0.0037; sign-flip p=0.96). It reduced F1 from 0.414 to 0.407 (difference −0.0064, 95% CI −0.0091 to −0.0039; p=0.0001) and recall from 0.418 to 0.404 (difference −0.0146, 95% CI −0.0214 to −0.0087; p<0.0001). At baseline triggers the spectral ratio did not separate marked-ripple triggers from triggers outside marks (median per-patient AUC 0.50). The reference was an automatic detector, not visual expert marks; a trigger outside a mark may still be a real ripple. Findings apply to this offline simulation, one reference band, a 20 ms envelope, and the sampled interictal records; hardware delay and stimulation artifacts were not modeled.

causal ripple trigger · spectral isolation · signal processing · SEEG · high-frequency oscillations · closed-loop stimulation · OpenNeuro ds003498