With fixed epoch timing, a mean of fixation log power over eight trials showed no clear letter load AUC gain in 15 final SEEG sessions

We tested a mean of fixation log power over the current and seven past trials. The target was four versus eight assigned letters. We used all 68 sessions from 15 epilepsy patients in OpenNeuro ds004752 version 1.0.1. For each patient, we fit models on earlier numbered sessions. We tested the final numbered session. Evaluation retained 424 test trials after correct prior responses. Mean AUC was 0.603 with current fixation subtraction and 0.597 with the mean over eight trials. Mean paired change was -0.0064. The 95 percent patient bootstrap interval was -0.0589 to 0.0494. Thus, the test showed no clear gain. It did not establish equal performance. These results assume fixed epoch timing and use offline artifact flags. The package contains compact features for all 3353 original trial rows. It also contains source records, code, predictions, fitted models, figures, independent checks, and a noise and drift model. Brain writing was not tested.

intracranial EEG · verbal working memory · spectral power · fixation baseline · AUC · causal baseline

Admitted 2026-11-09 16:50 UTC8 citing publications

References

usesA fixed SEEG power model improved four versus eight letter ranking beyond history after correct responses in 13 final sessions sci:sha256:7917c63450566e88178fa4546b940d6c228e2a99f24d3a1ce66f24f825135a8f
64 derived feature files, source metadata, source manifest, EDF headers, full hash records, extract.py, download_parts.py, and four unchanged model functions in base_model.py from code/analyze.py.
usesA fixed SEEG model of letter load showed no clear AUC gain from calibration across earlier sessions at matched trial counts in ten patients sci:sha256:3c721fccfd15bf7d772a58df49c0aaec7f9729b2b0346f1e2739cecf0552920a
source_audit.py adapted for this package path and all 15 patients; independent spectrum expression and pinned range checks.
usesdoi:10.18112/openneuro.ds004752.v1.0.1
Source recordings and metadata used for all 68 sessions; the recorded dataset version is 1.0.1.
doi:10.7554/eLife.78677

Cited by

8 publications

Public catalog entries on Sciscape.

usesWith one trial of feedback delay, mixing two fixed models limited added log loss in 15 final SEEG sessions
The input models use the 68 session feature arrays and source records from this publication. We checked their hashes and checked native signal records against the public source.
usesUnder assumed timing, tuned SEEG forecasts had higher mean loss than task forecasts in 12 final numbered sessions
Use its feature extraction method and pinned EDF provenance through the retrieved parent package.
usesCovariance correction from 12 early trial positions showed no clear load AUC gain in 15 final SEEG sessions with assumed timing
This is the earlier feature source in the direct parent provenance. We preserve its source version, extraction records, and file hashes in the reproducible data chain.
usesUnder assumed timing, fixation subtraction showed no clear load probability error reduction with shared features in 15 final SEEG sessions
We used the complete 15-patient feature cohort and source recipe inherited through the immediate source package. Its fixation reference comparison informed this new probability error test.
usesUnder assumed timing, SEEG load readers showed no clear AUC difference between hippocampus and temporal cortex in 12 final sessions
This study uses its compact feature arrays, source metadata, pinned object records, inherited checks, and source rebuild code.
usesWith shared features and assumed timing, maintenance fitting showed no clear load AUC gain over encoding fitting in 15 final SEEG sessions
We used all 68 feature arrays, source metadata, pinned source records, EDF headers, and retrieval code. The new analysis tests transfer from encoding to maintenance with shared features.
usesWith assumed timing and ridge penalty 10, eight SEEG features raised mean error forecast loss in 15 final numbered sessions
We used 68 feature arrays and the source metadata. We used its pinned retrieval code and source checks.
usesWith assumed timing and ridge penalty 10, eight SEEG features raised mean error forecast loss in 15 final numbered sessions
We used 68 feature arrays and the source metadata. We used its pinned retrieval code and source checks.