A fixed SEEG model of letter load showed no clear AUC gain from calibration across earlier sessions at matched trial counts in ten patients
We tested whether calibration trials from several earlier sessions improve final-session ranking of four versus eight letters. We used public SEEG data from OpenNeuro ds004752 version 1.0.1. Ten patients had at least three complete numbered sessions. The recent method used the session before the final session. The spread method used 100 draws from all earlier sessions. Each draw had the same trial count and class counts as the recent method. Each fit selected eight power features and used fixed ridge logistic regression. Trials followed a correct response and had no source artifact flag. There were 297 test trials. Mean AUC was 0.599 for recent trials and 0.587 for spread trials. The paired change was -0.012, with a patient bootstrap 95% interval from -0.062 to +0.037. The exact sign-flip p value was 0.654. Five patients gained AUC and five lost AUC. This result does not establish equal performance. Eight final sessions shared a date with the recent session. Session numbers set the test order. The test data were used in earlier work. The package contains figures, features, trial choices, fitted models, source hashes, code, and retrieval steps. Checks covered 300 public signal windows and all 2,040 fitted models.
stereoelectroencephalography · SEEG · calibration · working memory · AUC · brain-computer interface
References
3- usesA fixed SEEG power model improved four versus eight letter ranking beyond history after correct responses in 13 final sessions sci:sha256:7917c63450566e88178fa4546b940d6c228e2a99f24d3a1ce66f24f825135a8f
This work uses the exact frozen band-power features, source metadata, source hashes, and extraction code. It tests calibration across earlier sessions at matched trial and class counts in ten eligible patients.
- usesdoi:10.18112/openneuro.ds004752.v1.0.1
The analyzed SEEG recordings and task metadata came from OpenNeuro ds004752 version 1.0.1.
- supportsdoi:10.7554/eLife.78677
This paper describes the source task and its trial sequence, including the rule that after a wrong response the next load is four letters.
Cited by
7 publicationsPublic catalog entries on Sciscape.
- usesWith fixed epoch timing, a mean of fixation log power over eight trials showed no clear letter load AUC gain in 15 final SEEG sessions
source_audit.py adapted for this package path and all 15 patients; independent spectrum expression and pinned range checks.
- usesUnder assumed timing, tuned SEEG forecasts had higher mean loss than task forecasts in 12 final numbered sessions
Use its separate source-power audit method to check retained and fresh native EDF windows.
- usesCovariance correction from 12 early trial positions showed no clear load AUC gain in 15 final SEEG sessions with assumed timing
We used its raw range audit and complete EDF stream reconstruction methods through the direct parent. These methods checked source powers and complete raw file hashes in this investigation.
- usesUnder assumed timing, fixation subtraction showed no clear load probability error reduction with shared features in 15 final SEEG sessions
We used its independent EDF range audit method, inherited through the immediate source package. We repeated raw range checks in all 68 sessions.
- usesWith shared features and assumed timing, maintenance fitting showed no clear load AUC gain over encoding fitting in 15 final SEEG sessions
We used its independent power expression and pinned range audit method through the direct parent package. We ran fresh checks in all 68 sessions.
- usesWith assumed timing and ridge penalty 10, eight SEEG features raised mean error forecast loss in 15 final numbered sessions
We adapted the independent power check from this audit method through the parent package.
- usesWith assumed timing and ridge penalty 10, eight SEEG features raised mean error forecast loss in 15 final numbered sessions
We adapted the independent power check from this audit method through the parent package.