Separate penalties and a fallback left higher mean 50 ms offline theta phase error than own history across files from 36 SEEG patients
We tested whether other electrode shafts improve a phase forecast when their coefficients have a separate penalty. We used four contact pairs and two interictal files from each of 36 patients in OpenNeuro ds004100, version 1.1.3. We chose penalties in file A. The choices included an exact model that uses only the target pair history. The main test used file B, with no new fit. Mean cosine loss rose from 0.57912 to 0.58570. The added loss was 0.00657, with a 95% patient bootstrap interval of 0.00224 to 0.01159. Mean angular error rose from 58.72 degrees to 59.22 degrees. Separate penalties reduced error by 4.43 degrees compared with one shared penalty. The fallback was chosen for 62 of 144 pairs. It did not remove the small mean cost on the second file. This is an offline test on a reused epilepsy cohort. The phase target uses future voltage. The package contains selected source samples, source records, model files, forecasts, figures, and code. The result concerns this model and these selected pairs. It does not establish accuracy during brain stimulation.
SEEG · theta phase · phase forecasting · ridge regression · signal processing · brain-computer interfaces
References
5- usesPast theta phase gave no clear gain for 70–110 Hz power forecasts at a 50 ms delay in ridge models from 36 SEEG patients sci:sha256:e47f4cca1ddda495b2ea555a6d3df714855a541e13213ccdf073e8cb3c4b668a
Its cohort selection and native data provenance were inherited through the direct input publication. We preserved those source records. We did not use its power forecast outcomes to select the phase models.
- usesAdding three SEEG pairs increased mean 50 ms offline theta phase error after 56 s of ridge fitting in seven rest recordings sci:sha256:37e0a21c84e6afb5f766949c16aad144051ab4e2a2974728bf474e5bad706520
Its seven-patient phase forecast result motivated the question. We used its centered offline theta phase definition as a method reference. We tested separate own and donor penalties, with an exact own history fallback, on another dataset and another file.
- usesOther shafts increased mean 70–110 Hz forecast error after removal of a 60 Hz fit in ridge models transferred between files from 36 SEEG patients sci:sha256:1c253197c3da76fad6c67aa954cbc5cfc87533a0045fd578dc1683888ff54373
We used its 72 native digital input files, selected contact pairs, pinned source receipts, and channel metadata for the 36-patient phase forecast test. We adapted its source audit code and made new source requests.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
Source dataset for the selected SEEG samples; exact version 1.1.3 is recorded in the publication and its saved metadata.
- doi:10.1038/s41467-025-59417-7
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesComplex voltage targets reduced mean 50 ms offline theta phase error across files from 36 selected SEEG patients
We used its exact selected SEEG inputs, source records, and own-history model conventions. We changed the fitting target and tested fixed models on file B.