In seven SEEG patients, encoding power predicted recall (median AUC 0.686), while task-adjusted within-position AUC was not significant

This analysis tested whether SEEG power during word encoding predicted later recall after accounting for serial position and list number. Seven patients from OpenNeuro ds004789 v3.1.0 (session 0, 500 Hz) were analyzed with leave-one-list-out cross-validation. Neural features were log10 power at eight frequencies from 3 to 140 Hz, averaged over 0–1.6 s after word onset. Median neural AUC was 0.686 (6/7 patients above 0.5; exact Wilcoxon p = 0.031). After task regression, within-serial-position median AUC was 0.613 (6/7 above 0.5; p = 0.109). The neural AUC decrease after task regression and within-position scoring was positive in all seven patients (median 0.042; p = 0.016). The median fraction of above-chance neural AUC retained was 0.80 (pooled 0.68). Neural AUC exceeded task-only AUC by a median 0.019, a difference that was not significant (p = 0.58). A one-number spectral tilt measure decreased from median AUC 0.573 overall to 0.518 within serial position. Thus task structure explained a small, consistent part of neural classifier AUC, while the remaining within-position evidence was not significant at group level. The sample was a time-limited convenience subset of seven 500 Hz patients; 1000 Hz patients were not analyzed. There was no per-patient permutation test, and group tests had low power with n = 7.

SEEG · free recall · word-list memory · encoding power · serial position · leave-one-list-out cross-validation · area under the ROC curve · spectral tilt

Admitted 2026-11-09 09:07 UTC20 citing publications

References

usesdoi:10.18112/openneuro.ds004789.v3.1.0
Source dataset: Delayed Free Recall of Word Lists, OpenNeuro ds004789 v3.1.0. The publication used session 0, acq-bipolar EDF files; the cached exact-version metadata records this DOI and a CC0 license.

Cited by

20 publications

Public catalog entries on Sciscape.

usesA cross-patient SEEG decoder did not predict recall within serial position in 7 held-out RAM FR1 patients; patient-specific decoders did better
Uses the archived per-word SEEG band-power features of 7 RAM FR1 patients and the within-serial-position AUC method.
usesIn 18 FR1 patients, an 8-channel SEEG decoder retained 83% of above-chance AUC but scored below the all-channel decoder
Uses its within-serial-position AUC as a control for task-structure effects in the recall decoder.
usesIn 18 SEEG patients, word-n recall-decoder output weakly predicted next-word recall after serial-position control
The feature extraction code (fetch_features.py) and the features of 7 patients in c4b8c884 come from this publication.
usesA Gaussian kernel did not improve SEEG word recall prediction over a linear decoder in 18 patients
Feature-extraction code (fetch_features.py) and 7 of the 18 patient feature files come from this publication, through c4b8c884.
usesIn seven SEEG patients, an unvalidated k=8 IED detector found no significant group effects, but classifier associations differed by patient
We used the subject list (7 FR1 patients of ds004789), the Morlet log-power feature set and the classifier settings (L2 logistic regression, C = 7.2e-4) of this publication.
usesIn 18 SEEG patients, 3–40 Hz retained full-channel recall AUC, while four-channel decoding had lower AUC
The feature extraction and decoder settings (C = 7.2e-4) come from this publication.
supportsSEEG recall decoding adds to word memorability and serial position in 18 patients
That work showed that serial position and list effects explain part of recall decoding; we add word memorability as a second no-brain factor and show that neural data still add information beyond both.
usesEight selected bipolar channels retained all-channel word-recall decoding accuracy in 10 SEEG patients
Source A features (7 patients, Morlet band power, embedded electrodes.tsv atlas labels used for the region analysis) were extracted by this publication.
challengesIn 19 pyFR patients, recall decoding remained above chance within serial position (AUC 0.571; offline, one session each)
In FR1, within-position AUC was not significant in 7 patients; in 19 pyFR patients it is significant, so the serial-position explanation does not generalize to pyFR.
usesIn 10 SEEG patients, training ranked 8 channels matched all channels and beat random sets in leave-one-list-out recall decoding, but not across session halves
Source of the band-power features and atlas labels for 7 patients.
usesAcross 10 ds004789 FR1 SEEG patients, cross-patient recall decoding was weak and not significant within serial position
Per-word Morlet band power features of 7 ds004789 FR1 patients (session 0) and the within-patient regularization C.
usesIn 18 SEEG patients, recall decoding persisted weakly to the next word and two-word smoothing reduced AUC
The features that we analyzed were computed with the feature-extraction code of this publication (Morlet power at 8 frequencies, bipolar SEEG, OpenNeuro ds004789).
usesThe IED and recall-classifier protocol produced no data results because downloads stalled
Byte-range EDF reader, Morlet features and the 7-subject set.
usesAcross 18 SEEG patients at four sites, other-site training was non-inferior for word-recall decoding (median AUC difference +0.004; margin 0.02)
Its feature-extraction code made the band-power features (via c4b8c884) that this analysis uses.
supportsSEEG recall prediction replicates in pyFR, but frozen FR1 weights do not transfer
In 16 pyFR patients, the within-patient AUC decreased when computed within serial position (W_chan minus W_chan_withinSP +0.023, 12/16, p = 0.039), as d371cb2a found in 7 FR1 patients. Different from d371cb2a, the within-position AUC stayed significant here (0.600, 15/16, p = 4.6e-5), probably because n is larger.
supportsIn 18 FR1 SEEG patients, word memorability predicted recall but explained little of the session-0 decoder score
It found that encoding band power predicts recall and that serial position and list effects explain part of the AUC. We test a different stimulus factor, word memorability, and find that it explains almost none of the decoder AUC.
supportsSEEG band power predicts later free recall in ds004865, including after serial position is controlled
We also tested the serial-position confound. In ds004865 the decoder signal stays when the permutation keeps serial position fixed (median within-position AUC 0.578, p = 0.013).
challengesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder
It reported within-serial-position recall AUC (0.61) as not significant in 7 patients. With 18 patients, the same measure (0.618) is significant (p = 5.3e-5), so the earlier null result was a power limit.
usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder
We used its feature code without change and its 7 feature files (500 Hz patients), and we fetched its 12 remaining queued SEEG patients.
usesA fixed cross-patient recall prior did not improve patient SEEG decoding with 1–3 labeled lists
The features in c4b8c884 come partly from the feature code and files of this work.
usesIn 17 SEEG patients, calibration on the first eight lists showed no significant AUC loss versus time-blind training in one session
The features that we analyzed were computed with the feature-extraction code of this publication (Morlet power at 8 frequencies, bipolar SEEG, OpenNeuro ds004789).