In 19 pyFR patients, baseline-subtracted HFA was not above chance; an unplanned raw-HFA read-out was above chance from 0.6 s

We tested how soon a read-out with no trained parameters can separate later-recalled from not-recalled words after word onset. We analyzed archived per-word channel-mean 70–150 Hz power (high-frequency activity, HFA) from 19 pyFR SEEG sessions, one per patient. The planned score was mean HFA over [0, t] minus mean HFA over the 0.5 s before onset. It was not above chance at any planned decision time (t = 0.2–1.0 s; median AUC 0.516–0.529; exact sign-flip Holm p ≥ 0.78), or at 1.6 s (median AUC 0.522, p = 0.12). Pre-onset HFA alone did not predict recall (median AUC 0.520, p = 0.35). In an exploratory, unplanned test, raw HFA with no baseline was above chance from 0.6 s (median AUC 0.542, Holm p = 0.040) and at 1.0 s in 15 of 19 patients (median 0.528, Holm p = 0.015). A longer window to 2.1 s did not add AUC. The paired difference between the two scores was not significant (+0.038 AUC at 1.0 s, p = 0.19), so these data do not show that baseline subtraction removes recall information. This offline analysis used one session per patient, had weak effect sizes, and used a non-causal FFT band-pass with about 10–20 ms time spread.

high-frequency activity · HFA · intracranial EEG · SEEG · free recall · memory BCI · decision latency · area under the curve · sign-flip test