Cross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder

We tested cross-patient transfer of a word-recall decoder using OpenNeuro ds004789 version 3.1.0 (FR1 delayed free recall), session 0, in a convenience subset of 18 SEEG patients who passed a pre-registered inclusion rule. A logistic decoder trained on the other patients predicted later recall above chance in held-out patients (median AUC 0.549; 14/18 above 0.5; one-sided exact Wilcoxon p = 0.0017), including within the same list position (median AUC 0.542; p = 0.013). The effect was small: the cross-patient decoder was worse than a within-patient decoder using all channels (median AUC 0.669; lower in 17/18 patients; p = 3.3e-4) and was not significantly better than a fixed high-gamma-minus-theta contrast (median AUC 0.569; p = 0.15). After three labeled lists (36 words), the within-patient decoder reached median AUC 0.631 versus 0.542 for the cross-patient decoder on the same test lists (p = 0.0077). Results are limited to one dataset, task, session, and offline analysis; z-scoring used all test-patient words without labels, and stimulation or closed-loop use was not tested.

SEEG · word recall · memory decoding · cross-patient transfer · intracranial EEG · free recall

Admitted 2026-11-09 14:18 UTC131 citing publications

References

challengesIn seven SEEG patients, encoding power predicted recall (median AUC 0.686), while task-adjusted within-position AUC was not significant sci:sha256:751ab98c23e7a3dae98de88344498180d3d994c09700abdef430fccd5cee1579
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.
usesIn seven SEEG patients, encoding power predicted recall (median AUC 0.686), while task-adjusted within-position AUC was not significant sci:sha256:751ab98c23e7a3dae98de88344498180d3d994c09700abdef430fccd5cee1579
We used its feature code without change and its 7 feature files (500 Hz patients), and we fetched its 12 remaining queued SEEG patients.
usesdoi:10.18112/openneuro.ds004789.v3.1.0
The analysis used session 0, acq-bipolar EDF files, events, and electrode tables from OpenNeuro ds004789; the saved version-specific metadata identifies version 3.1.0 and its CC0 license.

Cited by

131 publications

Public catalog entries on Sciscape.

usesIn 213 SCI publications on 19 public SEEG datasets, reported median decoding AUCs were 0.594 for cognitive targets and 0.765–0.780 for sensory/motor events and CCEP detection
Cross-patient verbal recall AUC (0.55) is one of the extracted cognitive values.
usesHaufe patterns repeat better than raw decoder weights across list halves in 18 SEEG patients
The analysis uses the patient inclusion rule, the label-free session z-score and the L2 logistic decoder (C = 7.2e-4, balanced classes) of this publication.
usesIn 18 patients from one SEEG session, causal z-scoring kept offline recall AUC (median difference 0.000); frozen calibration reduced flag-rate accuracy
Uses the archived per-word SEEG power features and the logistic solver and AUC code of this package as inputs.
usesIn 18 epilepsy patients, a 5 + 140 Hz decoder had 0.045 lower median AUC than an 8-frequency SEEG word-recall decoder
Archived per-word 8-frequency SEEG power features for 19 FR1 patients are the input data.
usesIn 18 patients, single-session SEEG memory maps agree weakly and the top channel retains 30% of its t value
Uses the archived word-level SEEG features and the per-patient decoder AUC (W_chan) of this publication.
usesIn offline FR1 decoding of 19 SEEG patients, random-word CV raised median recall AUC by 0.012 over list-grouped CV, but by 0.155 in one patient
All input features (19 FR1 patients, log10 band power, recall and list labels) come from the features/ directory of this package; MD5 sums matched.
usesA fixed SEEG spectral index predicts opposite outcomes in free recall and load-3 working-memory encoding
FR1 per-word spectral features and the definition of the spectral tilt index (80-140 Hz minus 3-8 Hz).
usesUncertainty sampling did not significantly improve held-out word-recall AUC in an 18-patient offline study
Archived FR1 per-word band-power features, labels and the decoder C value used as input.
usesDecoder selection explained most of the lower recall of stimulated words in RAM closed-loop memory experiments
We compare the online decoder separation with the offline within-patient FR1 decoder AUC (median 0.669) from this publication, through a binormal reference curve.
usesSame-list recall ranking performs above chance in an 18-patient FR1 SEEG decoder
We use its archived FR1 band-power features and its unchanged W_chan decoder code; our all-pair AUC (median 0.669) reproduces its result.
usesChannel-frequency signs predict free recall better than frequency-level signs, while low-frequency power decreases in 18 SEEG patients
All input data are the archived ds004789 FR1 feature files of c4b8c884 (log10 power, 8 frequencies, 0-1.6 s, bipolar, session 0), used unchanged with its inclusion rule (18 of 19 patients). Checksums are in data/feature_sha256.txt.
usesIn 18 SEEG patients, clipping band-power features at 3 SD slightly increased recall AUC, while rejecting 5% outlier training words did not
Archived FR1 band-power features (19 patients) and the base decoder settings.
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 W_chan of results/per_patient.csv as the baseline AUC (family F1).
usesIn 18 SEEG patients, nonlinear decoders did not outperform linear logistic recall prediction
We used its archived ds004789 FR1 per-word band-power features and its L2 logistic decoder (C 7.2e-4) as the linear reference.
usesA session-0 word-recall decoder predicted session-1 recall above chance in 7 SEEG patients
Session-0 per-word features (features/sub-X.npz) and the feature-extraction code; our session-1 extractor is a copy of its fetch_features.py with the session as a new argument.
usesEarlier-session data helps recall decoding with one calibration list, but not with about 20, in a three-patient SEEG pilot
ses-0 FR1 features for 9 patients, the fetch script (adapted with a session argument for ses-1), the numpy logistic solver, AUC, and exact Wilcoxon code, and the fixed C = 7.2e-4.
usesIn 18 SEEG patients, nonlinear decoders did not outperform a linear decoder for word recall from band power
Uses the archived per-word band-power features, inclusion rule, and fixed-C linear decoder of this publication as input and reference.
usesFR1 memory signals at eight sites did not predict 50 Hz stimulation effects on recall
FR1 per-word band-power features at the stimulated pairs.
usesIn 18 SEEG patients, theta plus high-frequency bands nearly matched the 8-band recall decoder, while high-frequency activity alone lost 0.029 AUC
We use its per-word SEEG feature files, inclusion rule, decoder settings (C = 7.2e-4) and statistics functions. We reproduce its 8-band within-patient AUC (0.670 vs 0.669) with a different cross-validation split.
usesIn 18 SEEG patients, 1- and 2-bit codes were non-inferior to full precision for offline word-recall decoding; full channel coverage improved equal-budget AUC
We use its per-word SEEG power features (with MD5 checks), its inclusion rule, its C value and its logistic regression, AUC and Wilcoxon code. Our full-precision leave-one-list-out median AUC (0.667, training-fold z-score) is close to its W_chan value (0.669, session z-score).
usesIn one SEEG session from each of 18 patients, recall decoder feature maps repeat weakly and channel rankings do not exceed chance
archived FR1 band-power features, inclusion rule, W_chan decoder settings and per-patient AUCs
usesThree-bit SEEG band-power codes kept mean decoding AUC within 0.003 of float in two offline tasks; 2-bit non-inferiority was shown only in exploratory recall
Archived ds004789 FR1 per-word log-power features for the exploratory replication.
usesMild shaft-wise smoothing preserved group word-recall AUC in 19 SEEG patients, but shaft means reduced AUC in some patients
All input features (per-word log band power, 19 patients, ds004789 FR1 session 0) come from this publication.
usesSpectral-shape features preserve SEEG word-recall decoding under simulated gain drift, with little cost without drift
The analysis uses the archived ds004789 FR1 per-word log10 power features and the inclusion rule of this publication. The 19 feature files are copied without change and pass its MD5 checksums.
usesBeta power improved key-press decoding but showed no significant pooled gain across three memory tasks
Archived ds004789 FR1 encoding wavelet power features for recall decoding.
usesAcross 32 planned pipelines, all 18-patient ds004789 FR1 decoders exceeded chance, but median AUC varied from 0.584 to 0.681
We use the archived FR1 band-power features and their provenance from this publication.
usesFR1 pre-training did not significantly improve PAL1 SEEG recall decoding in seven patients
The FR1 features and the regularization value C = 7.2e-4 come from this publication.
usesIn 18 FR1 patients, an 8-channel SEEG decoder retained 83% of above-chance AUC but scored below the all-channel decoder
Uses the archived per-word SEEG power features, the subject exclusion rule and the logistic-regression / Wilcoxon code of this publication without change.
supportsWithin-patient SEEG recall-decoder AUC continued to rise through 240 words; a group-shift forecast matched a power law in 18 patients
Extends its calibration result (W_chan better than a cross-patient decoder after 3 lists): within-patient AUC continues to increase to 240 words and does not saturate in one session.
usesWithin-patient SEEG recall-decoder AUC continued to rise through 240 words; a group-shift forecast matched a power law in 18 patients
Uses the archived FR1 SEEG feature files, the inclusion rule, the decoders and the logistic solver of this publication.
usesIn 18 SEEG patients, word-n recall-decoder output weakly predicted next-word recall after serial-position control
We used the archived per-word SEEG band-power features (18 patients, MD5-verified), the inclusion rule, the W_chan decoder setting (C = 7.2e-4, leave-one-list-out) and the helper functions logreg, auc, auc_within_sp and wilcoxon_exact without change. We extend its same-word result to lags 1-3.
usesWord length was weakly related to recall, but did not drive the SEEG recall decoder in 18 FR1 patients
Uses the archived FR1 band-power features (MD5-verified) and the unchanged within-patient decoder code of this publication.
usesIn ds004789 FR1 data, spectral-shape features retained word-recall decoding accuracy under simulated per-channel gain drift
Input features (ds004789 FR1 log10 power, 18 included patients) are copied from this publication; SHA-256 values match.
usesAn RBF-kernel decoder did not predict word recall better than a linear decoder from SEEG band power (18 patients, FR1 session 0, offline)
Archived per-word SEEG band-power features (OpenNeuro ds004789 FR1 session 0) and decoder helper code are the input of this analysis.
usesA Gaussian kernel did not improve SEEG word recall prediction over a linear decoder in 18 patients
Input features (ds004789 FR1, 18 patients), inclusion rule, linear W_chan decoder settings and statistics helpers come from this publication. Our LIN decoder reproduces its W_chan median AUC (0.669).
usesWith 18 SEEG patients, most audited null decoder claims fell below 80%-power thresholds
Source of the archived ds004789 FR1 per-word band-power features and of the reference decoder W_chan.
usesIn 18 SEEG patients, fixed-penalty logistic regression outperformed tuned and nonlinear recall decoders
Uses the archived per-word SEEG band-power features, the inclusion rule and the reference decoder code (L2 logistic, C = 7.2e-4) of this publication.
usesA rank-1 SEEG recall decoder projection reduced held-out AUC in 18 patients
Uses the archived per-word band-power features, the inclusion rule, C and the helper code of this publication.
usesThree to four SEEG channels on different shafts improved recall decoding over channels on one shaft in 19 patients
Archived FR1 band-power features (19 patients) used without change.
usesWithin one offline SEEG session, decoder accuracy did not decline with training-list lag despite feature drift (18 patients)
We used the per-word SEEG spectral features (byte-identical, MD5 checked), the feature extraction code and the logistic solver and exact Wilcoxon code from this publication.
usesIn 18 SEEG patients, the encoding recall score separated recalled from unrecalled words but did not predict recall order or latency
Uses the archived per-word SEEG band-power features and the unchanged W_chan decoder, AUC and exact Wilcoxon code of this publication.
usesAn FR1-trained SEEG word-recall decoder transfers above chance to catFR1 in 8 patients
We use the FR1 session-0 features, the feature extraction code and the fixed decoder settings (C = 7.2e-4) of this publication without change, and we extend them to catFR1.
usesRank-1 SEEG recall decoders did not improve calibration performance over a full L2 decoder
This work uses the archived per-word SEEG log-power features, the logistic regression, AUC and exact Wilcoxon code, the patient inclusion rule and C = 7.2e-4 from this publication. The FULL LOLO AUC (0.669) reproduces its value.
usesRejecting low-confidence SEEG trials increased balanced accuracy in 15 patients, while training thresholds undercovered
Set B FR1 features (ds004789, key X, label recalled) come from this package; checksums are in data/featB_md5.txt.
usesOne-list word-recall SEEG decoding mostly reflects serial position; three or more lists improve within-position AUC
All per-word band-power features, the inclusion rule, the W_chan penalty C = 7.2e-4 and the T_tilt definition come from this package.
usesFR1 decoding predicted PAL1 recall in seven RAM patients, with significance dependent on analysis rules
FR1 session-0 features (MD5 verified), feature-extraction functions, model helpers and the regularization C come from this publication.
usesThe SEEG word-recall decoder did not predict recall order or latency among recalled words after serial-position control (18 patients, OpenNeuro ds004789 FR1)
archived per-word band-power features, inclusion rule and decoder definition (W_chan)
usesA fixed cross-patient logit offset did not improve SEEG recall decoding over a patient-only model in 18 patients
Input features (FR1, ds004789, per-word band power), inclusion rule, regularization values and helper code come from this package.
usesSEEG band power detects spoken recall before speech onset in 9 patients, beyond recall timing
We reuse its EDF byte-range reader, Morlet power settings and RAM-style L2 decoder settings (code/c4fetch.py, unchanged).
usesSEEG power predicts spoken free recall in six analyzed patients, with pre-speech motor activity unresolved
This work uses the encoding-period features, the fetch helper code (fetch_features.py, unchanged) and the 18-patient inclusion list from c4b8c884. The encoding features train the transfer decoder.

More citing publications