SEEG power predicts spoken free recall in six analyzed patients, with pre-speech motor activity unresolved

We tested whether stereo-EEG (SEEG) power distinguishes windows before spoken free recall from silent windows during a word-list recall period. The pre-registered primary window was 0.9–0.1 s before an isolated recalled-word onset. We analyzed six of 18 planned patients from OpenNeuro ds004789, version 3.1.0, using Morlet log power at eight frequencies (3–140 Hz) and leave-one-list-out logistic regression. The median AUC was 0.852 (6/6 patients above 0.5; one-sided exact Wilcoxon p = 0.016), and the median AUC within five-second bins of time since recall start was 0.831 (6/6; p = 0.016). Time since recall start alone had median AUC 0.724. An earlier window, 1.6–0.8 s before speech, had median AUC 0.710. An encoding-period decoder transferred to retrieval with median AUC 0.685 (5/6 patients above 0.5). Power at 13–22 Hz decreased before recall. The remaining 12 planned patients were not analyzed because downloads were slow; the fixed queue order may have caused selection bias. The late window may contain pre-speech motor, breath, or muscle activity, so the result does not isolate a memory signal. The analysis was offline and used one session per patient and all bipolar channels.

stereo-EEG · free recall · memory · neural decoding · power spectrum · AUC · speech motor activity