In 9 patients, a cross-patient threshold cut seizure-detector false alarms from 98.7 to 11.5 per hour as sensitivity fell from 100% to 66%

In a leave-one-patient-out test of a causal line-length detector for stereo-electroencephalography (SEEG), a cross-patient threshold reduced pooled pre-ictal false alarms from 98.7 to 11.5 per hour across 9 HUP patients and 35 clipped seizures. Sensitivity fell from 35/35 (100%) to 23/35 (66%). False alarms were lower in 8 of 9 patients (median decrease 73 per hour; two-sided Wilcoxon p = 0.039). A cross-patient L2 logistic model did not improve window ranking over the max-z score (mean AUC 0.872 vs 0.873; p = 1.0), and the max-z score with the same cross-patient threshold rule had similar results (22/35 detected; 10.5 false alarms per hour). A smoothed score with a threshold from each patient’s 5-minute interictal clip produced 222.5 false alarms per hour. These results suggest that the reduction came from the threshold source, not the learned model. The dataset contained only 0.95 hours of pre-ictal data; the reported false-alarm rates are not long-term estimates. The study used one feature, one model type, and one target rate in a small sample.

stereo-electroencephalography · seizure detection · line length · false alarms · cross-patient threshold · leave-one-patient-out · responsive neurostimulation