In 13 seizure-free patients, one 5-minute fast-ripple map ranked the resected zone nearly as well as a 30-minute map, but maps differed across nights

Closed-loop SEEG devices must select targets from short interictal recordings. We measured how many minutes of NREM sleep are necessary before a high-frequency oscillation (HFO) map ranks the resected zone (RZ), and if the map is stable across nights. We used the dataset authors' HFO markings for all 385 NREM intervals (about 5 min each; 58 nights; 20 patients) of OpenNeuro ds003498, and RZ labels from the clinical sheet (publication 9059ecb3). The plan was written before the analysis. Primary result: in the 13 seizure-free patients, the fast-ripple (fr) map from 6 random intervals did not rank the RZ significantly better than the map from 1 interval (per-patient AUC median 0.74 to 0.78; paired median difference +0.029, 10 of 13 positive, exact Wilcoxon p=0.092). The top-1 fr pair of one interval was in the RZ with probability 0.89 (0.92 with 12 intervals). But it was the same pair as the top-1 pair of the held-out recording only with probability 0.69 (0.86 with 6 intervals). Two intervals from the same night gave more similar maps than two intervals from different nights (median Spearman rho 0.63 vs 0.51; 11 of 12 patients; p=0.0015). In the 7 patients who were not seizure-free, more recording did not make the map point to the RZ. Thus, 5 min is sufficient to find an RZ-like region, but selecting one specific stimulation pair needs more intervals and more than one night. Limits: one center; RZ is the label, not the seizure-onset zone; author markings were not re-detected; the pair set is the union of pairs with any marked event; p-values are not corrected except for the one primary test.

high-frequency oscillations · fast ripples · stereo-EEG · sleep · resected zone · epilepsy · target selection · recording duration · night-to-night reliability