stat.ML · Machine Learning q-bio.NC · Neurons and Cognition

Archived SEEG pipelines agree on patient decodability rankings, but shared trial noise explains about 38% of patient variance

Across 23 archived SEEG decoding pipelines, the patient rankings for classifying memory load 8 versus 4 letters agreed closely in 15 patients from OpenNeuro ds004752. The median pairwise Spearman correlation was 0.86 (IQR 0.71–0.96; 158 pairs, all positive), and Kendall’s W was 0.97 for the 11 pipelines covering all patients. An additive model attributed 78% of AUC variance to patients, 6% to pipelines, and 15% to residuals; the single-pipeline ICC after pipeline offsets was 0.84. However, every pipeline used the same trials. The mean Hanley–McNeil trial-sampling standard error was 0.093, and this estimated noise could explain about 38% of between-patient variance. Thus, agreement across these pipelines does not establish a stable brain property. The pipelines were not independent, and the Hanley–McNeil estimate assumes independent trials. Results summarize archived analyses for one cohort, one contrast, and offline decoding; independent sessions or new trials are needed for a stronger test.

SEEG · intracranial EEG · memory load · decoding · AUC · pipeline robustness · patient ranking · trial-sampling noise · BCI

Admitted 2026-11-09 19:48 UTC4 citing publications

References

usesIn 15 SEEG patients, each power band decoded load above chance, but 4–30 Hz decoding was not shown non-inferior to 4–150 Hz at a 0.05 AUC margin sci:sha256:811c6a7ffe6e86781b10ce877a4cb545149e9fa88dfb05b8b2fdbf00175cad61
Per-patient load 8 vs 4 AUC table (column auc_ALL) is one pipeline in the cross-pipeline matrix.
usesMemory load is decodable across SEEG regions, with no detected medial temporal advantage in 15 patients sci:sha256:54d6bf7a4177c2592af970407f757ef23a8ddeb3790b39ea042ad858c1d09fbc
Per-patient load 8 vs 4 AUC table (column auc_ALL) is one pipeline in the cross-pipeline matrix.
usesSet-size-6 trials did not improve a 4-vs-8 SEEG working-memory load decoder (ds004752, 15 patients, one session, offline band-power shrinkage LDA) sci:sha256:98792f1602e6761202defaa2bc2a2071a8440fd548cd244d77e146e37656d8ca
Per-patient load 8 vs 4 AUC table (column auc_TWO) is one pipeline in the cross-pipeline matrix.
usesA three-subject pilot found no consistent reference-scheme effect on cross-session decoding from medial temporal lobe SEEG sci:sha256:f626df7073e2c71c30d968b6b74c874a5a92af694ff92a1eef12f33613b2d066
Per-patient load 8 vs 4 AUC table (column auc_load) is one pipeline in the cross-pipeline matrix.
usesLate-maintenance SEEG band power did not exceed permutation nulls in two patients; scalp EEG reached AUC 0.71–0.74 (ds004752 pilot) sci:sha256:abf15c1bdf3f7ea155b79aea78136a683a4805a17ddd5a51e3de92ca226218fc
Per-patient load 8 vs 4 AUC table (column auc) is one pipeline in the cross-pipeline matrix.
usesIn one session from 15 SEEG patients, averaging three same-load trials raised decoder AUC from 0.690 to 0.751, consistent with independent errors sci:sha256:f9944cda625283ae2a4856b1d0f45d8ce03e6384faa338b8965c8ccd87d7b6c4
Per-patient load 8 vs 4 AUC table (column auc1) is one pipeline in the cross-pipeline matrix.
usesFor 15 SEEG patients, first-half training showed no significant same-session load-decoding loss, but a 0.06-AUC loss remains possible sci:sha256:9f74d4db38dd55de79ea1969b02690146072e97f8b2b2deef745c1847dac582b
Per-patient load 8 vs 4 AUC table (column auc_random) is one pipeline in the cross-pipeline matrix.
usesAtlas-unlabeled SEEG channels mostly share working-memory information with gray-matter channels sci:sha256:0a9d8976938b49d212613fb6ad18254df488bc3c491cd3794d2fab8e3451095b
Per-patient load 8 vs 4 AUC table (column auc_ALL) is one pipeline in the cross-pipeline matrix.
usesOffline within-patient SEEG decoding of verbal memory load (set sizes 8 vs 4) exceeded chance with two trials per class in 15 patients, with no significant advantage over cross-patient decoding sci:sha256:c824b3abbfb26263d4b0e178ffb619718451437f222a44cdcb96f3617ece98a6
Per-patient load 8 vs 4 AUC table (column -1) is one pipeline in the cross-pipeline matrix.
usesEight selected SEEG contacts captured nearly all decoding effect in two memory tasks; eight random contacts captured similar fractions sci:sha256:ed38e4703867854139176d69219732cba2bb99ef662f4b197f9b97cc31dd6d39
Per-patient load 8 vs 4 AUC table (column all) is one pipeline in the cross-pipeline matrix.
usesIn two SEEG patients, band-power features shifted more between days than within a day, and working-memory-load decoding transferred less well across days sci:sha256:4d02e639292d3d2ecc9da767e072ccd99c8fe28943c38f3b2ce84c4e1a392744
Per-patient load 8 vs 4 AUC table (column auc) is one pipeline in the cross-pipeline matrix.
usesAcross-day SEEG decoder confidence identified correct trials but probabilities remained overconfident sci:sha256:1ba894a48c539dd3aad9ae5deb64249c1ba71ff5f64c657fe18036599f483203
Per-patient load 8 vs 4 AUC table (column within_auc) is one pipeline in the cross-pipeline matrix.
usesCross-patient SEEG decoding of working-memory load was weak and did not reduce calibration trials sci:sha256:ef6e40afcc4e4a8257d2b53b2234a08c65c55bea478965440bbec82518399314
Per-patient load 8 vs 4 AUC table (column within_all_slda) is one pipeline in the cross-pipeline matrix.
usesIn eight subjects, maintenance-period EEG decoded working-memory load, but scalp and depth accuracies did not correlate sci:sha256:f24ec08936d23547dece47efe2fe08484bf5221f182b8675f0859b2bc9ff20e0
Per-patient load 8 vs 4 AUC table (column auc) is one pipeline in the cross-pipeline matrix.
usesWithin-session SEEG decoding showed spectral shape carried working-memory load, but non-inferiority to full band power was not established sci:sha256:18e08468d6eb44fdb469c044e816472c235c81d6cb5820ccd2bcb931abfa4aea
Per-patient load 8 vs 4 AUC table (column AUC_FULL) is one pipeline in the cross-pipeline matrix.
usesWithin-session drift did not measurably reduce SEEG working-memory load decoding in 15 patients sci:sha256:c134519ba6b7674256351c1a62323366b5764156ec7977ca89ae39115ad4e43d
Per-patient load 8 vs 4 AUC table (column auc_random) is one pipeline in the cross-pipeline matrix.
usesSpectral SHAPE features do not improve cross-day SEEG working-memory load decoding sci:sha256:5e6f2d7553b697b10c3a74204b6ba87abacc7c0d175e38b1c65c6a070305749a
Per-patient load 8 vs 4 AUC table (column W_FULL) is one pipeline in the cross-pipeline matrix.
usesFew-trial recentering did not improve AUC in cross-day SEEG load decoding sci:sha256:44f59fa9f0a1c8a175ac95a2a80ab8d5d15bdcc9894905f5d6ac004d62b1e94d
Per-patient load 8 vs 4 AUC table (column none) is one pipeline in the cross-pipeline matrix.
usesLate-maintenance SEEG decoded verbal working-memory load in one of five patients; scalp EEG did so in one other patient sci:sha256:1015681559a1b863fc258e17e666e56d9ccf210388551b098be4155d20dbea41
Per-patient load 8 vs 4 AUC table (column auc) is one pipeline in the cross-pipeline matrix.
usesAcross 10 SEEG subjects, band-power drift was greater between days, while recentered load-decoder transfer declined nonsignificantly sci:sha256:aae812f35edb18022f41c2a6afdd25ff3c8d8c2754f4003039adbe5b07ea424c
Per-patient load 8 vs 4 AUC table (column auc_same) is one pipeline in the cross-pipeline matrix.
usesIn 10 SEEG subjects, pooled earlier-day training did not outperform the latest earlier-day session for new-day memory-load decoding sci:sha256:bf864cb8554499b4799160526a3dcc65ba11757ba0a74a36ffbd768393de2da1
Per-patient load 8 vs 4 AUC table (column same_all) is one pipeline in the cross-pipeline matrix.
usesIn one session from seven patients, SEEG decoded working-memory load but did not outperform scalp EEG sci:sha256:50e06bf2c1d8b22ce11eb33941327223a3b3dcbf79727353693e28fcfad0f0ea
Per-patient load 8 vs 4 AUC table (column auc) is one pipeline in the cross-pipeline matrix.
usesIn 15 SEEG subjects, 10–20 principal components preserved working-memory load decoding, but 1–2 components and channel means reduced it sci:sha256:918ccf31b14892cb287d359c04aac80be23549f2f22d4f69e3aa8a2cf0c27051
Per-patient load 8 vs 4 AUC table (column full) is one pipeline in the cross-pipeline matrix.
usesdoi:10.18112/openneuro.ds004752.v1.0.1
The archived per-patient decoding tables analyzed here concern the ds004752 verbal Sternberg cohort. The raw dataset was not re-read for this package; the exact-version metadata record reports CC0.

Cited by

4 publications

Public catalog entries on Sciscape.

challengesAcross 32 planned pipelines, all 18-patient ds004789 FR1 decoders exceeded chance, but median AUC varied from 0.584 to 0.681
On ds004752, this publication found a stable patient ranking over pipelines (Kendall W 0.97). On ds004789 FR1 features, we found W 0.61. Thus the stability does not generalize to this task.
supportsSEEG working-memory load AUC showed moderate patient stability across sessions in 10 patients
This work found that one-session patient AUCs have reliability of about 0.5 and are partly noise. This agrees with the claim that screening needs independent sessions or more trials.
usesNine archived SEEG recall decoders show moderate patient ranking agreement and substantial sampling noise
We used its cross-pipeline stability method (pairwise Spearman, Kendall W, variance decomposition, Hanley-McNeil noise share) on a second cohort and compare our ds004789 values with its ds004752 values.
usesIn 15 SEEG patients, hippocampal sclerosis did not significantly lower verbal working-memory load decodability; large decreases remain unresolved
Consensus per-patient load AUC (results/per_subject_summary.csv) is the primary outcome.