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

1. 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.

2. 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.

3. 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.

Public catalog entries on Sciscape.

• usesRBF kernel ridge decoding showed no detectable recall gain over a linear kernel in 18 patients with offline SEEG band power

  Uses its archived per-word SEEG features, inclusion rule, AUC and exact Wilcoxon functions without change, and its per-patient W_chan AUCs for a secondary comparison.

• usesIn 18 SEEG patients, median AUC rose from 0.535 to 0.549 with 1–17 training patients, with no detected gain from 8 to 17

  Input per-word SEEG band-power features, cross-patient model definition and helper code; our N = 17 result reproduces its X_cross AUC (0.549, 14/18, p = 0.0017).

• usesShrinkage LDA did not reproduce a spatial-residual advantage in 18 patients decoding word recall

  We used its archived ds004789 FR1 feature files and inclusion rule without change.

• supportsNonlinear and tuned decoders do not predict later word recall from SEEG band power better than a fixed L2 logistic decoder (18 FR1 patients, one session, 5-fold list-grouped CV)

  With 5-fold list-grouped CV the W_chan decoder gives a median AUC of 0.670 (17/18 above 0.5). This agrees with the published leave-one-list-out value of 0.669.

• usesNonlinear and tuned decoders do not predict later word recall from SEEG band power better than a fixed L2 logistic decoder (18 FR1 patients, one session, 5-fold list-grouped CV)

  We use its archived per-word SEEG band-power features, its inclusion rule and its fixed L2 logistic decoder (W_chan) as the reference.

• usesCoarse MNI-region features did not improve a cross-patient SEEG word-recall decoder in 18 patients

  Input band-power feature files and electrode tables (ds004789 FR1, 18 patients); tests the open limit that channel-mean features discard location.

• usesIn 18 SEEG patients, 3–40 Hz retained full-channel recall AUC, while four-channel decoding had lower AUC

  We use its archived per-word band-power features of ds004789 FR1 and copy its decoder, AUC and exact Wilcoxon functions.

• usesRandom training-label errors reduced word-recall AUC in an 18-patient offline SEEG study; the effective-sample rule held at 10% and 20%, not shown at 30%

  We used its archived FR1 band-power feature files, its inclusion rule, its W_chan decoder settings and its kernel logistic solver.

• usesFive pre-calibration patient properties did not predict recall-decoder AUC in 18 SEEG patients

  Uses the archived FR1 features, electrode tables and per-patient W_chan AUCs of this package as input.

• usesLeft and right SEEG channels decoded later word recall similarly in eight patients

  We used the archived ds004789 FR1 encoding features and the decoder functions from this publication.

• usesIn 5 FR1 patients, undetected simulated channel failure reduced word-recall AUC more than detected failure

  We use the archived FR1 SEEG features (ds004789) and the decoder solver of this publication.

• usesSEEG recall decoding adds to word memorability and serial position in 18 patients

  Uses its per-word SEEG band-power features and its W_chan decoder setup; our replication gave the same median AUC (0.669).

• usesSummed per-shaft SEEG scores were non-inferior for word-recall decoding, with a small AUC loss (ds004789 FR1, 18 patients)

  Original source of the per-word band-power features, the inclusion rule, and the regularization C = 7.2e-4.

• usesEqual-count SEEG decoding found no medial temporal advantage over lateral temporal pairs in 18 patients

  Input data: archived per-word band-power features of 18 SEEG patients (ds004789 FR1, session 0), the inclusion rule and the within-patient decoder settings (C = 7.2e-4).

• usesIn 13 patients, channel-resolved SEEG recall AUC rose through 20 lists within one FR1 session

  Uses the archived per-word band-power features, the decoder settings (C values) and the decoder and statistics code.

• usesIn 18 SEEG patients, recall decoding improved with each added bipolar channel; selection helped only at 1 to 2 channels

  Uses the archived per-word SEEG power features and the W_chan decoder of this publication; the all-channel AUCs reproduce its W_chan result (median 0.669, 17 of 18 above 0.5).

• usesIn 18 SEEG patients, channel count, MTL channel count, and recall rate showed no large monotonic relation with recall-decoder AUC

  We use its per-patient W_chan AUCs and its feature files (channel lists and electrodes.tsv atlas labels) without change.

• usesIn a four-patient RAM PAL1 pilot, study-phase SEEG band power did not predict cued recall consistently

  We applied the fixed theta/high-frequency index idea from FR1 to a cued-recall task as comparator index B.

• usesSpreading eight SEEG pairs across shafts modestly improved word-recall decoding, but not within list position

  We used its archived per-word band-power feature files (ds004789 FR1, 18 patients), its inclusion rule, and its logistic-regression solver, AUC and exact Wilcoxon code.

• usesSimulated noisy SEEG channels slightly reduced recall-decoder AUC; repair did not improve the primary AUC test

  We used its archived ds004789 FR1 feature files, its inclusion rule, its W_chan decoder settings (C = 7.2e-4) and its solver, AUC and Wilcoxon code without change.

• usesIn 18 FR1 SEEG patients, 10% simulated random label flips lowered median held-out AUC by 0.020; 20% exceeded the effective-sample rule

  Input features (ds004789 FR1 spectral power per word) and inclusion data.

• usesRecall decoding from SEEG increased with each added shaft up to the full implant in 18 patients

  We used the archived per-word log-power features (features/sub-*.npz, ds004789 FR1 session 0), the electrodes.tsv text in each file for shaft labels, and the reference logistic, AUC and exact Wilcoxon code from this package.

• usesMath distractor performance did not predict list recall in 18 SEEG patients; a held-out decoder showed a small association

  We used the archived per-word band-power features (features/*.npz, including the events.tsv text with the math PROB rows) and the decoder code (code/analyze.py lines 1-104: logistic regression, C = 7.2e-4, Wilcoxon) of this publication. The included patients and the inclusion rule are the same.

• usesIn 18 SEEG patients, encoding-power PC1 did not predict word recall and underperformed a fixed frequency tilt

  Archived per-word band-power features of ds004789 FR1, helper code, and the fixed tilt readout.

• usesThe SEEG word-recall decoder had similar low-score and high-score tail separation in 18 patients

  Uses the archived per-word band-power features and the within-patient decoder settings of this publication.

• supportsZero-calibration SEEG key-press decoding was near chance in 8 subjects, while 10 labeled within-subject trials reached median AUC 0.83

  Also finds that cross-subject SEEG decoding without channel alignment is weak and not better than within-subject decoding, here for a motor task.

• usesCoarse region features did not improve zero-calibration cross-patient SEEG recall decoding in 18 FR1 patients

  Uses the archived FR1 power features, the X_cross decoder and the test code of this publication; answers its open question about region-aware cross-patient decoders.

• usesAt matched pair counts, SEEG word-recall decoding showed no significant MTL advantage over lateral temporal cortex in 18 patients

  Per-word band-power features, electrode tables, source ETags and decoder helper code come from this package.

• usesIn 18 patients, a Platt map improved calibration of SEEG word-recall probabilities, but Brier-skill gains were small

  We used the archived per-word SEEG band-power features and the W_chan decoder definition (C = 7.2e-4, class-balanced) from this publication.

• usesLow-frequency SEEG features retained recall-decoding performance in 18 patients; high-gamma-only features performed worse

  We use its archived band-power features, patient selection and W_chan decoder code without change.

• usesIn 18 patients, selecting 8 SEEG pairs did not improve recall decoding over random pairs

  Per-word band-power features (ds004789 FR1, session 0) and decoder, AUC and Wilcoxon helper code.

• usesIn 18 SEEG patients, recall decoding persisted weakly to the next word and two-word smoothing reduced AUC

  We used the archived per-word log-power features (19 npz files, MD5 checked), the inclusion rule, the W_chan decoder (C = 7.2e-4) and the T_tilt score of this publication, and its helper functions without change. Its own-word within-position AUC (0.618) is our lag-0 reference.

• usesAn adaptive threshold reduces stimulation-rate error in an offline memory-decoder simulation

  Uses the archived ds004789 FR1 band-power features, the decoder solver, and the penalty C = 7.2e-4 from this package.

• usesSpread SEEG channel placement improved memory decoding in three computable tests

  Archived FR1 (ds004789) band-power features for 19 subjects.

• usesPre-CV feature selection inflated SEEG word-recall AUC and produced high AUCs with permuted labels

  We use its archived per-word FR1 band-power features as input. They are copied in data/ and MD5-verified.

• usesA session-0 SEEG recall decoder predicted session-1 recall above chance in seven patients

  Uses its archived FR1 session-0 features, its feature-extraction code (session made an argument) and its W_chan decoder code without change.

• usesFixed L2 penalty outperforms per-patient tuning for SEEG recall decoding in 18 FR1 patients

  Uses its 19 per-patient FR1 SEEG feature files (MD5 checked) and its model code; reproduces its W_chan leave-one-list-out median AUC 0.669.

• usesForgotten words that later returned as intrusions had forgotten-like SEEG decoder scores

  We use the archived FR1 band-power features and events.tsv text of this package (checksums in results/source_features_MD5SUMS.txt).

• usesIn 18 SEEG patients, two-frequency word-recall decoders were not non-inferior to an eight-frequency decoder

  Archived per-word band-power features, provenance and inclusion rule.

• usesIn 18 SEEG patients, an equal-weight cross-patient ensemble did not improve few-list word-recall decoding

  Uses the encoding-power features, inclusion rule, X_cross design and numpy model and statistics code of this publication.

• usesA session-0 SEEG recall decoder predicted session-1 recall above chance in four patients

  We used its feature code (with a session argument), its analysis functions without change, and its 4 session-0 feature files (same MD5). Our 3-list pooled calibration result (0.644) agrees with its 3-list result (0.631).

• 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)

  We used its archived per-patient band-power feature files, its inclusion rule, its X_cross model and its numpy solver, AUC and Wilcoxon code without change.

• usesIn 18 SEEG patients, training without primacy words lowered mid-list recall-decoder AUC (median difference -0.019, p = 0.027)

  Uses the archived per-word FR1 band-power features, inclusion rule and W_chan decoder functions of this package; reproduces its within-position AUC (0.618).

• supportsPast-only normalization kept cross-patient SEEG recall decoding above chance in 18 patients

  Its stated limit was non-causal test-patient z-scoring. With past-only or first-list statistics the median AUC stayed 0.540-0.543 (p <= 0.005), so that limit did not inflate its result.

• usesPast-only normalization kept cross-patient SEEG recall decoding above chance in 18 patients

  We used its archived feature files and its cross-patient decoder code (copied without change) and changed only the normalization of the test patient.

• usesSEEG band-power patterns drift in 19 of 19 patients, but did not predict next-word recall beyond lag in 9 patients

  We use its 19 word-level SEEG band-power feature files and the events.tsv text stored in them, without change.

• usesRemoving features above 70 Hz changed AUC little in two offline SEEG tasks; non-inferiority held only in ds004789

  ds004789 FR1 Morlet features (19 patients), NCYC = 5 for band edges, OpenNeuro provenance

• usesFour to eight screening lists improved 8-channel recall decoding over random channels in 18 SEEG patients, but performance remained below all channels

  Input per-word band-power features (19 npz files) and the numpy logistic solver come from this package.

• supportsCalibration-free SEEG recall read-outs predict free recall in 19 pyFR patients, with median AUCs of 0.551 and 0.531

  The calibration-free read-outs of this work also predict recall in an independent cohort (ds004865) with no loss of T_tilt effect size.

• usesCalibration-free SEEG recall read-outs predict free recall in 19 pyFR patients, with median AUCs of 0.551 and 0.531

  Feature code, FR1 training features, T_tilt and X_cross definitions, and FR1 per-patient AUCs.

More citing publications

First page of citations