Training across assumed uniform 0 to 100 ms delays reduced mean offline theta cosine loss in seven SEEG rest records
We tested one public SEEG rest record from each of seven people. Each record supplied four selected bipolar pairs. We assumed a known uniform command delay from 0 to about 100 ms, independent of the signal. A ridge reader trained on the mean unit phase across these delays had lower later mean cosine loss than a reader trained at the mean 50 ms delay. Loss fell from 0.8119 to 0.7925. All seven person means improved. The descriptive 95% person bootstrap interval for the loss decrease was 0.0136 to 0.0250. Mean absolute phase error remained 74.95 degrees. An exact vector calculation gave a full-future cosine ceiling of 0.4240 and a timing loss floor of 0.5760 for one forecast at every tested cutoff. Gains were near zero at narrower delay ranges. A model trained for broad delays had higher loss when scored at a fixed delay. The package contains all input voltage, models, scores, source versions and hashes, code, a reader application function, and three figures. Fresh source retrieval matched all 3.36 million selected values. Independent checks passed. This is an exploratory offline method result. It does not show live timing accuracy or a stimulation benefit.
SEEG · theta phase · phase prediction · command delay · ridge regression · brain-computer interface · cosine loss
References
4- usesA learned causal filter reduced mean 50 ms offline theta phase error compared with a calibrated sinusoid in seven SEEG rest records sci:sha256:eee4ee9691c2f7bdc7469e4d6d40001adfb8ef71b4c7d2c8e1992f7d6b193a1d
This publication supplied the exact seven bipolar input arrays, fixed source versions and hashes, source retrieval code, and numerical filter operations. We adapted its phase reference, features, and ridge fit to test training across an assumed delay range.
- usesEarly frequency selection reduced mean 50 ms offline theta phase error by 3.55 degrees in seven SEEG rest recordings sci:sha256:5c3ae075960de0774e0fcc176ee07a5e3c9a353d86238f1e49023a23aa4d2bfa
The selected pairs and original source preparation came from this publication through publication 81b6655a-8291-4575-a114-c55a3c38b8d2. We preserved and repeated the early pair selection and kept its exact selected voltage.
- usesdoi:10.18112/openneuro.ds006065.v1.0.0
This is the exact OpenNeuro dataset version used for the seven source rest records; the package preserves source metadata, pinned object versions, hashes, and selected voltage.
- doi:10.1038/s41467-025-59417-7
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesFits to eleven requested 25–75 ms delays reduced mean offline theta phase loss across files from 36 SEEG patients under an independent delay law
Its reviewed result supplied earlier method credit and evidence context for the delay-average unit target and known-trajectory loss bound. It found a gain under broad assumed delays in seven ds006065 rest records, but not at its moderate range. The current study extends this method to another dataset and 36-patient separate-file tests. No numerical file or code was imported from it.
- usesA max-min phase rule had higher mean worst loss than an early selected forecast for two delay laws in seven SEEG rest records
Uses the seven preserved four-pair SEEG voltage products, pinned range records, source retrieval code, and forecast math. Adds a test of an unknown mixture of two delay laws and an exact max-min phase rule.