In 18 ds004789 FR1 SEEG patients, previous-word decoder output weakly predicted next-word recall, while averaging consecutive outputs reduced recall ranking

We used the leave-one-list-out W_chan SEEG decoder of c4b8c884 (0-1.6 s power features) on 18 patients of OpenNeuro ds004789 FR1. We tested, with a pre-registered plan, if the decoder output of word n-1 predicts recall of word n at the same serial position. The median AUC was 0.530 (11/18 patients above 0.5, exact Wilcoxon one-sided p = 0.037). Only 2/18 patients had a list-permutation p below 0.05. When we removed the list mean of the output, the effect went away (0.516, p = 0.37). Thus the information is a slow list-level state. A decoder trained on the features of word n-1 did not do better (difference -0.015, p = 0.06). Lags 2 to 4 gave AUC near 0.5. The average of the outputs of words n and n-1 decreased the AUC within serial position by 0.022 (Holm p = 0.034). The average of three words decreased it by 0.038 (Holm p = 0.0026). Behavioral recall of word n depended strongly on recall of word n-1 (+0.17). Exploratory result: the output computed in the window of word n+1 (2.5-4.1 s after onset of word n) predicted recall of word n (AUC 0.574, 17/18). This stayed after control for recall of word n+1 (0.553, p = 0.0024). For closed-loop memory stimulation with this decoder, use the output of the current word alone. Limits: one session, one linear decoder, 2.5 s word interval, offline outputs.

SEEG · free recall · memory decoding · serial position · AUC · closed-loop brain stimulation · word recall

Admitted 2026-11-09 16:58 UTC3 citing publications

References

Cross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
[Legacy relation: reproduces] With its features and code, the leave-one-list-out W_chan decoder gives the same median AUC (0.6685 here, 0.669 reported) over the same 18 patients.
usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
We use its archived ds004789 FR1 features, inclusion rule, W_chan decoder and helper code without change. We reproduce its lag-0 result (median AUC 0.669).
usesdoi:10.18112/openneuro.ds004789.v3.1.0
The analysis used session 0 bipolar SEEG data from OpenNeuro ds004789 v3.1.0 (FR1).

Cited by

3 publications

Public catalog entries on Sciscape.

supportsSame-list recall ranking performs above chance in an 18-patient FR1 SEEG decoder
It found that the lag-1 decoder signal is a slow list-level state; we show that this list-level state is small and that the decoder AUC for the current word does not depend on it.
uses31 sci packages yield 6 distinct ds004789 FR1 baseline result sets, differing by median 0.044–0.066 AUC per patient across 17 patients
We used column auc0_raw of results/per_patient.csv as the baseline AUC (family F1).
supportsIn 18 SEEG patients, within-list recall decoding had median AUC 0.659 vs 0.669 overall; list-centering did not improve it
Our decomposition shows that the list-level part of the same-word decoder output carries little recall information (list-mean AUC 0.519). This agrees with its finding that the small next-word prediction is list level.