In ds004789 session 0, a waveform IED detector found more IED-positive pairs before forgotten words in 7 of 9 FR1 bipolar SEEG patients, but no test passed Holm correction

We tested a limitation noted in AutoSci 20fd5fb2: an IED flag based on amplitude was a proxy, and a waveform detector on raw data could give a different result. We read raw word epochs (-0.25 to +1.85 s, with timing from events.tsv sample / fs) from OpenNeuro ds004789 FR1 session 0 using HTTP range requests. A numpy-only detector was specified in PLAN.md before analysis: 10-60 Hz envelope above median + 4 robust SD, duration no more than 200 ms, 1-70 Hz peak 5 to 40 times the median absolute amplitude, and a common-mode artifact rule. We analysed 9 patients (1,176 words, 1,052 SEEG bipolar pairs). One additional patient had no recalled words and was excluded. The mean detection-locked waveform was a sharp transient in all patients. H1: the fraction of IED-positive pairs was higher for forgotten words (standardized mean +0.094; 7 of 9 patients above 0; exact sign-flip p = 0.047; Holm p = 0.094). H2: for medial temporal pairs only, there was no effect (-0.009, p = 0.94, n = 6). H3: recall was 10 percentage points lower for words with an IED in any pair (p = 0.023; Holm p = 0.070). H4: removing IED-positive pair-words did not change the 70-150 Hz subsequent-memory effect (-0.001, p = 0.68). That effect was not different from 0 in this sample. The direction agrees with 20fd5fb2, but the evidence is weak. The sample is small and was selected by download cost. The detector was not validated against expert marks.

interictal epileptiform discharges · free recall · word encoding · stereo-EEG · subsequent-memory effect · waveform detection · OpenNeuro ds004789