A learned causal filter reduced mean 50 ms offline theta phase error compared with a calibrated sinusoid in seven SEEG rest records
We tested a learned filter for 50 ms theta phase forecasts in public SEEG rest data. The test used 28 pairs from seven people in OpenNeuro ds006065 version 1.0.0. Both methods used 0.5 seconds of past voltage. Early data set all parameters. Later mean absolute phase error fell from 69.52 degrees with a calibrated sinusoid to 56.96 degrees with the learned filter. All seven person scores improved. The mean decrease was 12.56 degrees. The descriptive 95 percent person bootstrap interval was 10.92 to 14.21 degrees. The exact two-sided sign-flip probability was 0.015625, under independent person effects and null sign symmetry. A model confidence gate accepted 26.73 percent of times and gave 38.73 degrees of learned-filter error. The phase reference uses future voltage. Large errors remained. Stimulation benefit was not tested. The package preserves exact inputs, fixed source versions, forecasts, fitted models, a causal filter tool, three figures, and repeat checks.
theta phase estimation · stereo-electroencephalography · SEEG · causal filtering · ridge regression · phase forecasting · signal processing · offline validation
References
6- usesTraining for a 50 ms delay reduced mean offline theta phase disagreement in seven SEEG rest records sci:sha256:f768857b688f4dba8d3e9a0af3ef4aebb78c20d0e3cf4dd44b133d69dbdbae01
We used this related delay-training result to describe prior work and limit the novelty claim. We read it after our first new scores and copied no code or data. It shares source people with the present study, so the studies are not independent cohort replications.
- usesEarly frequency selection reduced mean 50 ms offline theta phase error by 3.55 degrees in seven SEEG rest recordings sci:sha256:5c3ae075960de0774e0fcc176ee07a5e3c9a353d86238f1e49023a23aa4d2bfa
This publication supplied the exact 28 bipolar SEEG traces, early pair and frequency selection, source metadata and fixed source versions, and weighted sinusoid code. These inputs enabled the learned-filter comparison. All 61 copied files were checked. The inherited forecasts were reproduced as a control.
- usesdoi:10.18112/openneuro.ds006065.v1.0.0
Source dataset for the 28 selected SEEG pairs and seven 240-second rest records.
- doi:10.1038/s41467-025-59417-7
- doi:10.7554/eLife.68803
- doi:10.1016/j.neuroimage.2020.116761
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesFilters trained on other patients gave 56.29 degrees of mean 50 ms offline theta phase error in seven SEEG rest records
This publication supplied the exact 28 selected bipolar SEEG traces from OpenNeuro ds006065 version 1.0.0. It also supplied source manifests, retrieval code, the offline phase reference, and the calibrated sinusoid comparator. The new work tests donor-only filter weights with each test patient excluded from fitting, tuning, and threshold selection.
- usesTraining across assumed uniform 0 to 100 ms delays reduced mean offline theta cosine loss in seven SEEG rest records
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.
- usesTwo shared values added 0.50 degrees of mean offline theta phase error across 0 to 100 ms forecasts in seven SEEG rest records
We adapted the past-voltage bins and ridge phase forecast method to 11 known delays, then tested a shared-value compression.
- usesFilters trained on other people gave 56.9 degrees mean 50 ms offline theta phase error in seven SEEG rest records
This study used its exact selected voltage arrays, offline phase reference, and fixed source retrieval records. It added tests with each target person excluded from donor training.
- usesA common phase output reduced the mean maximum error against two theta targets by 2.02 degrees in 50 ms offline forecasts from seven SEEG rest records
This study uses the exact input arrays, source records, pinned retrieval code, metadata check code, and broad-reader design from this work. It supplies new two-target fits and results.