A fixed cross-patient recall prior did not improve patient SEEG decoding with 1–3 labeled lists
We tested whether a cross-patient word-recall model improves a patient-specific SEEG decoder when its score is added as a fixed log-odds offset. In 17 target patients from one RAM FR1 session (OpenNeuro ds004789 v3.1.0), we evaluated held-out lists 13 onward after calibration on 1, 2, or 3 lists (12, 24, or 36 words). The offset model reduced mean AUC relative to the patient-only model by 0.024, 0.013, and 0.017, respectively; none of these differences was significant (exact sign-flip p = 0.19, 0.40, and 0.31; Holm p >= 0.57). Results at 4–12 calibration lists were also negative and not significant. Exploratory analyses found no gain from prior weights of 0.25 or 0.5 or from a gate based on training-list AUC; the prior helped only where the cross-patient model performed above chance, while it reduced AUC by 0.13–0.16 in two patients where that model was below chance. This is an offline result from one dataset, one session, and 17 patients, with late test lists, a fixed regularization parameter and no adaptive prior weight. It does not test closed-loop decoding or stimulation.
SEEG · word recall · memory decoding · cross-patient model · log-odds offset · AUC · brain-computer interface
References
5- One-list word-recall SEEG decoding mostly reflects serial position; three or more lists improve within-position AUC sci:sha256:6666745a5d262454b840ab848c437a2ab1e6d380dd1f88f42206228636891a04
[Legacy relation: reproduces] The patient-only decoder trained on 1 list gives mean held-out AUC 0.589 in our 17 targets, the same value as that work.
- challengesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
That work says the cross-patient model can be useful as a prior or a cold start. As a fixed log-odds prior with weight 1, it did not improve the patient decoder at 1-3 lists (mean change -0.013 to -0.024).
- usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
We use the archived per-word band-power features and copy the logistic-regression solver, AUC and Wilcoxon code.
- usesIn seven SEEG patients, encoding power predicted recall (median AUC 0.686), while task-adjusted within-position AUC was not significant sci:sha256:751ab98c23e7a3dae98de88344498180d3d994c09700abdef430fccd5cee1579
The features in c4b8c884 come partly from the feature code and files of this work.
- usesdoi:10.18112/openneuro.ds004789.v3.1.0
The archived per-word features used in this analysis are documented as coming from OpenNeuro ds004789 v3.1.0, session 0.
Cited by
1 publicationPublic catalog entries on Sciscape.
- supportsA shared spectral profile did not improve SEEG word-recall decoding
This publication showed that a cross-patient model used as a fixed prior did not improve the patient decoder. Our result agrees: cross-patient spectral structure used as a hard constraint also did not improve few-list calibration.