Median power lowers excess crossings near marked HFOs at matched calibration rates in eight sleep SEEG records

We tested whether a median of five short high-gamma power values reduces crossings near marked high-frequency oscillations (HFOs). We used OpenNeuro ds003498 version 1.1.1, with one sleep SEEG record from each of eight patients with epilepsy. The primary test used 88 exact-name bipolar pairs. Each method used a separate threshold from the first 120 s. Both had a 5% calibration crossing rate. On later data, mean power had a marked-minus-unmarked crossing excess of 14.85 percentage points. Median power had 7.46. The paired difference was -7.39 percentage points, with a 95% patient bootstrap interval from -11.95 to -2.80. Seven of eight patients had a negative difference. Median power increased unmarked crossings from 3.66% to 4.50%. The total test crossing rate changed little. A paired sine input test also showed a loss of response to short inputs. At fixed amplitude, 100 ms inputs added 63.30 percentage points for mean power and 17.70 for median power. These results show a feature tradeoff. The marks came from an automatic detector, and the data contain no intent or stimulation outcome labels. The package contains figures, code, power values, voltage samples, source hashes, a raw retrieval recipe, and reproduction instructions.

stereo-electroencephalography · high-frequency oscillations · high-gamma power · median filter · brain-computer interface · event detection