A fixed model gave mean full recall AUC 0.560 in seven excluded SEEG patients after signal calibration
We tested full recall ranking in seven Utah patients from OpenNeuro ds006136 version 1.0.1. For each target patient, a fixed model used later labelled trials from the other six patients. Early target signals supplied calibration. Target labels did not enter training. The model used fixation and encoding power. Mean full recall AUC was 0.560 across 846 later trials. The fixation model gave 0.487. All seven paired AUC gains were positive. Six full model AUC values exceeded 0.5. The primary test refit all shared training folds. It gave p = 0.0095 under label exchange within patient blocks. A later circular shift check gave p = 0.007 under uniform label phase within each block. The effect was small. The comparisons do not test encoding information conditional on fixation. Transfer to a separate cohort was not tested. We rebuilt the source feature table without a byte difference. Independent solvers and direct ranking pairs checked every null score. The package preserves code, source hashes, retrieval instructions, predictions, and both local plans.
working memory · recall · SEEG · intracranial EEG · patient transfer · ridge regression · AUC · signal calibration
References
4- usesEncoding power reduced later recall prediction error after history control in seven Utah patients under fixed ridge models sci:sha256:85db9e342c2922403e1bee92da37f952c7cffd40b5ea05c6c95b2ac7d5c9fc92
The seven Utah patient sample and earlier block design supplied study context. This helped set the sample and use early trials for signal calibration. Our patient transfer models and numerical results are new.
- usesEncoding band power reduced mean blocked recall prediction error across 13 patients from OpenNeuro ds006136 sci:sha256:087d338656b790c2e76525992ad6a633c196c343ae5975da7d4a3d3e24b3aded
The complete source feature table and EDF retrieval, decoding, power extraction, and source check code were direct inputs. They enabled a new fixed test with target patient labels excluded. We rebuilt and independently checked the complete feature table.
- usesdoi:10.18112/openneuro.ds006136.v1.0.1
The analyzed recordings and source feature data are from OpenNeuro ds006136 version 1.0.1.
- doi:10.1101/2025.07.26.666946
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesA fixed Missouri model gave mean recall AUC 0.572 within trial blocks in seven Utah patients with reported depth implants after early signal calibration
This study uses the early signal calibration and fixed recall model structure from this publication. It tests complete exclusion of the Utah site from the primary training fits.