In RAM FR3 and catFR3, decoder flag fraction did not predict unflagged-word recall across lists under a no-spillover assumption
We tested whether an online memory decoder used in RAM free recall (FR3) and categorized free recall (catFR3) reflected a slow state shared across a list, in addition to its per-word trigger. The decoder read SEEG or ECoG band power (8 Morlet frequencies, 3–180 Hz) during the 0–1366 ms interval after each word; output below the training median triggered a 500 ms stimulation train. We treated stimulated words as flagged and analyzed event and electrode files only (no iEEG signals). The analyzed files comprise 40 sessions; 38 sessions from 22 subjects met the requirement of at least 6 stimulation and 6 no-stimulation lists, yielding 399 stimulation lists. For each stimulation list, we related the flag fraction to residual recall, defined against recall at the same serial position in no-stimulation lists from the same session. Per-subject partial Spearman correlations removed session means and a linear list-number trend per session. The flag fraction did not predict recall of unflagged words (22 subjects; mean rho −0.049, 95% CI −0.170 to +0.073; sign-flip p = 0.45; within-session permutation p = 0.42); the SEEG subgroup was similar (16 subjects; rho −0.016, CI −0.165 to +0.130; p = 0.84; permutation p = 0.81). It also did not predict recall of flagged words (rho −0.045, CI −0.165 to +0.082; p = 0.50). Whole-list recall decreased as flag fraction increased (all subjects: rho −0.195, CI −0.324 to −0.057; p = 0.012; SEEG: rho −0.152, CI −0.300 to +0.010; p = 0.084), consistent with a composition effect from more low-recall flagged words rather than a detectable list-level decoder state. The inference about unflagged words assumes stimulation does not change recall of other words in the list. No classifier outputs were available, flag fraction is a coarse state read-out, and list estimates from 12 words are noisy; the confidence interval does not exclude a small effect (about rho −0.17 or weaker).
closed-loop brain-computer interface · memory · free recall · intracranial EEG · stimulation · decoder · list-level state · recall
References
3- usesOpen-loop stimulation during word encoding reduced recall by 1.6 points overall and 4.3 points at hippocampal sites sci:sha256:3a50433ef645749679239857437f277c2d17f9ae681ef4ff329341dc5afe1650
Earlier analysis of the same closed-loop datasets at the list level (no change in list recall).
- usesDecoder selection explained most of the lower recall of stimulated words in RAM closed-loop memory experiments sci:sha256:1140d5dd535ed3c590adc964507e2d119f41a2d0b5a0b96c7981524ac0741bfb
Same FR3/catFR3 data and flag definition; that work measured item-level decoder selection, this work tests a list-level state and uses its no-spillover finding as an assumption.
- usesIn 28-subject RAM FR2 depth-anode data, encoding stimulation reduced recall of stimulated words but showed no detectable change in the next word sci:sha256:b4af655bb7f17d9f2ba9a4128ba59cce7749bebb1dbca1d45d9a73da05b42ccd
Open-loop evidence that stimulation does not change recall of the next unstimulated word, which supports the no-spillover assumption.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesIn 3 FR3 SEEG sessions, 70-150 Hz power recovered within 0.2 s, while 3-8 Hz differences lasted up to 1.5 s after 0.5 s stimulation (2 subjects; zero-phase filters)
Uses the same closed-loop datasets and SEEG subject list. That work analyzed only events. This work adds neural signal analysis.