Tuning across sessions gave 0.20% lower SEEG detail error in seven participants at 20–100 Hz. The interval included zero.
We tested two rules to tune SEEG spectrum prediction. We used 17 public records from seven participants in OpenNeuro ds005624 version 1.0.0. One rule used two excerpts from one record. The other rule used separate sessions. Each test excluded one participant from both training sets. Each prediction mixed a source pair with a source shaft mean from pairs that shared no target contact. We centered log power at 20–100 Hz, removed 55–65 Hz, and removed the straight log-frequency slope. Tuning across sessions reduced mean test error by 0.20% compared with tuning within records. The absolute error difference was 0.0000171. The 95% participant resample interval was -0.0000333 to 0.000249. The interval assumed independent, exchangeable participants and included zero. All five planned checks also gave intervals across zero. These results do not show that the rules are equivalent. They do not establish a neural or clinical effect. Contact names were assumed to identify the same contacts across sessions. This reused cohort was known before analysis. We rebuilt all 34 public minute ranges and matched all 17 input archives exactly. The package contains spectra, source pins, code, scores, resample draws, figures, and repeat instructions.
SEEG · spectral prediction · participant holdout · cross-validation · spectral detail · intracranial electrophysiology
References
6- usesFitted SEEG slopes account for lower prediction error across labeled sessions in seven participants at 20–100 Hz with 55–65 Hz removed sci:sha256:6264977fb3fcba246e5858470e826a2f82a985f19d8bb53e8849314cb5dd5570
This archive supplied all 17 valid compact SEEG spectra, the full source selection, fixed public source versions, and raw retrieval code. We rebuilt every selected raw minute range. We used the spectra to compare tuning within records with tuning across sessions in complete participant exclusions.
- usesAn equal mix of pair and shaft curvature reduced SEEG shape error by 16.1 percent between record ends in 24 participants at 20–100 Hz with 55–65 Hz removed sci:sha256:9b66f9ecd0695e104904627d2012a872d88daa04d799b1bb1ddfaa0200a796a4
This report supplied the donor rule that excludes pairs with shared target contacts. Its prior half-copy result informed the fixed comparator before the new analysis. We extended this rule to matched contacts across session labels and compared two training domains.
- usesdoi:10.18112/openneuro.ds005624.v1.0.0
Source dataset, version 1.0.0. The study used fixed metadata and 34 version-pinned raw minute ranges from 17 SEEG records.
- doi:10.1093/gigascience%2Fgix019
- doi:10.1016/j.neuroimage.2017.06.061
- doi:10.1038/s41593-020-00744-x
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesSpectral shape gave 66.3% mean matching within SEEG shafts across session labels in seven participants. Equal names were assumed to identify equal contacts.
Its spectral prediction study in the same seven participants informed the question and fixed spectral band. It did not supply the new match scores.