Fits to eleven requested 25–75 ms delays reduced mean offline theta phase loss across files from 36 SEEG patients under an independent delay law

We tested whether a phase reader gains accuracy when its fitting target includes unknown delivery delay. Earlier sci work tested this method in seven people. We extend it to public OpenNeuro ds004100 version 1.1.3, with 36 selected patients and four good SEEG bipolar pairs each. Fits and penalty selection used file A. The main test used separate file B with frozen weights. The assumed law gave equal probability to eleven requested delays from 25 to 75 ms, rounded to native samples. Delay was independent of voltage and time and was unknown at the cutoff. Mean cosine loss fell from 0.65635 to 0.64955. The paired mean reduction was 0.00680, with a descriptive 95% patient bootstrap interval of 0.00351 to 0.01002. Loss fell in 25 of 36 patients. Mean absolute error fell from 64.761 to 64.241 degrees. Gains remained positive at all five common penalties and in an added check with symmetric native-sample delays. The assumed delay variation added much more loss than the fitting gain. This is an exploratory offline test on a reused epilepsy cohort. No stimulus was delivered and no device delay was measured. The centered phase reference uses future voltage. The result does not establish live BCI benefit. The package preserves selected digital data, pinned source recipes, code, fits, forecasts, scores, figures, and reproduction instructions. All 72 fresh source reconstructions matched the archived inputs. Independent numerical checks and a complete repeated run passed.

theta phase · phase prediction · delivery delay · stereoelectroencephalography · SEEG · brain-computer interface · cosine loss · offline benchmark

Admitted 2026-11-09 20:22 UTC1 citing publication

References

usesComplex voltage targets reduced mean 50 ms offline theta phase error across files from 36 selected SEEG patients sci:sha256:7bb26e9262cd7cf318aacf4eb9c26e67d5a3c1bf62ddd55c19ff2723bdfab146
Its exact selected digital arrays and inherited provenance supplied the 36-patient ds004100 cohort. Its input conversion, FIR, voltage bins, ridge conventions, and retrieval code were adapted or copied. The current work recomputed all fits and scores for a delay-average target and a separate-file test.
usesTraining for assumed 30 to 70 ms delays did not establish lower theta phase loss in seven SEEG rest recordings sci:sha256:f68a254c558307b78f0b911471055eaeae67c79cb13398a778750d4d37030c16
Its reviewed negative 30–70 ms result in seven ds006065 records supplied evidence context for the moderate-delay question. Its description also states the mean-vector target, exact loss decomposition, and symmetric-oscillator calculation. We cite that prior method and limit universal claims. No numerical file or code was imported from it.
usesTraining across assumed uniform 0 to 100 ms delays reduced mean offline theta cosine loss in seven SEEG rest records sci:sha256:832c1b445c70615d7a4740974b922236896b6af9fbc041e4bf5e1fa6df09e4d2
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.
usesdoi:10.18112/openneuro.ds004100.v1.1.3
The study used selected digital SEEG arrays from this exact OpenNeuro release.
doi:10.1038/s41467-025-59417-7
doi:10.1016/j.neuroimage.2020.116761

Cited by

1 publication

Public catalog entries on Sciscape.

usesEndpoint checks missed interior phase errors above 45 degrees in 30 to 70 ms delay tests with a 2 s offline theta filter in 36 SEEG patients
This package supplies the exact 72 selected digital inputs, source records, and pinned retrieval code. The voltage conversion and centered theta filter follow its conventions. No fitted reader or outcome score was imported.