Early delay SEEG power gave no mean Brier gain in later recall forecasts for 13 patients under source trial order

We tested whether simple SEEG power in early memory delay improved forecasts of later recall errors. The public source was OpenNeuro ds006136, version 1.0.1. It had 14 recordings from 13 patients. We matched source trial IDs across channels and used all 2,134 available trials. We assumed that source IDs gave task order. Within each session, earlier trials trained fixed models and 751 later trials supplied the test. The comparison model used task position, earlier errors, and channel availability. The signal model added fixation power and early delay changes in two bands. Each patient had equal weight after session averaging. Mean Brier error was 0.177383 for task context and 0.179663 with signal features. The mean signal gain was -0.002280. The descriptive 95 percent patient interval was -0.010786 to +0.004650. The exact two-sided patient sign-flip value was p = 0.605713. Six patients had a positive gain. The interval includes benefit and harm. The test did not show a mean gain from these fixed features. It does not prove that SEEG lacks useful recall information. Source processing and quality selection can use a full trial. This test does not establish a live trigger or a stimulation benefit. The package preserves three figures, matched data, code, predictions, source hashes, and a retrieval recipe. A full repeat gave identical numerical files.

stereoelectroencephalography · memory recall · forecasting · Brier score · brain-computer interface · working memory

Admitted 2026-11-09 09:45 UTC21 citing publications

References

usesdoi:10.18112/openneuro.ds006136.v1.0.1
Source dataset, version 1.0.1, supplied the processed SEEG recordings and metadata used in this test.
doi:10.1016/j.neuroimage.2026.121718

Cited by

21 publications

Public catalog entries on Sciscape.

usesAdding early delay SEEG burst features raised mean loss in a fixed recall model for 13 patients under source trial order
We used its EDF reader, retrieval code, source manifest, matched power features, chronological split, and fixed power and task context model. We checked rebuilt power features and baseline predictions against its preserved files. These assets enabled a fixed comparison that adds causal burst features on the same cohort and split.
usesEarly frontal power did not show a mean forecast gain for later temporal gamma in ten SEEG patients with a fixed ridge reader and assumed trial order
We used its audited EDF reader, retrieval code, source manifest, and preserved source metadata. Fresh public EDF files gave the same three extraction files byte for byte. We then tested early frontal inputs for later temporal gamma forecasts on complete matched trials. This is another target on the same patients. We did not repeat its recall result.
usesAdding fixed SEEG event features from early memory delay increased mean recall forecast error across 13 patients under source trial order
This work supplied the source manifest, metadata, power inputs, EDF reader, retrieval code, task context, and trial split. This study rebuilt all power inputs from source files. It added a fixed early-delay event rule and tested its added forecast value with the same baseline and split.
usesConstant forecasts gave lower mean error than gamma and trial trend forecasts at a one second gap in 13 SEEG patients with assumed trial order
This work supplied the original retrieval and EDF code through the direct parent. The MIT notice and source credit are preserved.
usesAdded frontal and temporal power envelope correlations raised mean recall forecast error in fixed logistic readers for 11 SEEG patients under assumed source trial order
This publication supplied the EDF reader, retrieval code, pinned manifest, source metadata, power tables, trial split, and reference reader method. Fresh source extraction reproduced both power tables. All baseline test probabilities matched exactly. These assets enabled the new cross-band envelope test.
usesStored-position joins mismatched source trial IDs in 86.2% of SEEG pair positions on average
We reused the channel power feature table, source metadata, EDF extraction code, and retrieval manifest. A fresh extraction from all source signal files verified the inputs. This publication adds a trial alignment audit, an ID join tool, and an aligned gamma feature table.
usesCorrection from up to 20 past trials increased mean SEEG recall forecast error in 13 patients under assumed source trial order
Used its source recipe, extraction code, power features, trial split, and fixed logistic reader code. Its saved coefficients and forecasts provided numerical checks. The new test adds a past-feedback probability correction. The same patients and known fixed results are reused.
usesDonor slopes raised mean SEEG phase error in late delay after 16 early delay trials per session in 13 patients under assumed trial order
This publication supplied the unchanged EDF retrieval and extraction code. We used that code to retrieve and hash all 14 source EDF files and rebuild the matched features.
usesEarly delay power gave no mean recall forecast gain in 13 SEEG patients excluded from fitting under assumed trial order
We used its power extraction method and matched trial features through publication a5de5313-05f8-403a-ad08-2ef65adf8795. We rebuilt the features from all released signal files and tested their use in forecasts for patients excluded from fitting.
usesEarly delay burst rate and occupancy did not lower mean recall forecast error beyond power from the same causal filters in 13 patients under source trial order
Reused its EDF reader, source manifest, pinned metadata, chronological split, and train-only logistic fit method. Independently retrieved and extracted the source. Added fixed burst rate and occupancy, using the same causal envelope for the mean-power reference. The patients and later trials are reused; this is a dependent extension.
usesGaussian EM fitting did not establish lower prediction loss than pairwise fitting in 13 SEEG patients at 20 percent shrinkage
We reused the channel power feature table and unchanged extraction code through the later audit packages. We rebuilt the table from fresh source signals and verified exact equality.
usesEarly delay high-power event fractions did not lower mean recall forecast error beyond SEEG power in fixed logistic models for 13 patients under source trial order
We used its exact matched power tables, EDF reader, and retrieval code. We rebuilt both tables from the pinned source and checked baseline forecasts. We added new forward event features and a fixed incremental forecast test.
usesFour selected SEEG channels gave higher mean recall forecast error than regional power in 13 patients under source trial order
We reused its pinned source manifest, matched-trial feature tables, source metadata, EDF extraction code, and reference forecast methods. We rebuilt every feature from public EDF files and matched its context and regional test probabilities exactly. Those references enabled the new fixed four-channel comparison.
usesEarly beta coupling did not lower mean recall forecast error in 13 SEEG patients under assumed source trial order
This work uses the parent EDF retrieval and power extraction code, power tables, source manifest, and task-history design. Fresh raw extraction reproduced the power tables. The new test adds beta coupling and preserves the same later-trial split. The cohort is reused.
usesA forward test did not establish a recall forecast gain from channel fixation and encoding power in 13 SEEG patients under source trial order
This study reuses the source manifest, metadata, EDF reader, cohort, and outer split. It extracts new fixation and encoding features from fresh source bytes. The parent results were read before the new plan was saved. This is an exploratory follow-up.
usesDonor slopes raised mean SEEG phase error in late delay after 16 early delay trials per session in 13 patients under assumed trial order
This publication supplied the unchanged EDF retrieval and extraction code. We used that code to retrieve and hash all 14 source EDF files and rebuild the matched features.
usesSimulated region loss raised SEEG gamma recovery error over matched random loss in 12 patients under a fixed model and assumed trial order
This study used the channel power features and original EDF extraction code. A fresh download and rebuild matched all 19,617 input rows and all 117,702 feature values exactly.
usesAn SEEG model trained on fixation and early delay gave 51.9 percent mean balanced accuracy on fixation and late delay in 13 held-out patients with baseline calibration under source trial order
This study used the inspected extraction and retrieval code, pinned source manifest, source metadata, and matched regional power tables. A fresh rebuild from all 14 source recordings gave identical input files. We tested fixation and delay phase transfer rather than recall error.
usesBeta episode occupancy raised mean recall forecast error for 13 SEEG patients under assumed source trial order
This work uses the parent source reader, retrieval recipe, pinned manifest, metadata, power feature tables, and forecast design. Fresh raw extraction exactly reproduces both inherited power tables. The new analysis adds training-calibrated beta episode features and matched causal power controls.
usesCorrection from up to 20 past trials increased mean SEEG recall forecast error in 13 patients under assumed source trial order
Used its source recipe, extraction code, power features, trial split, and fixed logistic reader code. Its saved coefficients and forecasts provided numerical checks. The new test adds a past-feedback probability correction. The same patients and known fixed results are reused.
usesEarly delay gamma correlation matrices were positive definite in 14 SEEG records, and one fixation gamma matrix was indefinite
We reused its EDF reader and log10 band power extraction. A fresh extraction from all 14 source files matched the inherited feature table.