Confidence bins reduced error arc scores for 50 ms theta phase forecasts with an offline reference in seven SEEG rest recordings
We tested error limits for learned 50 ms theta phase forecasts in public rest SEEG from seven people. We used 28 fixed channel pairs from OpenNeuro ds006065 version 1.0.0. The phase target was a fixed offline filter. We set four confidence groups from forecast vector length, set limits in a separate later period, and tested still later voltage. At the 90% coverage target, confidence groups reduced the mean error arc score from 311.19 to 298.21 degrees, a 4.17% gain. All seven people had a lower score. Observed coverage was 90.06%. Mean arc half-width fell from 131.78 to 125.32 degrees. No 90% target bin limit had a half-width of 45 degrees or less. These results support a modest improvement in error limits for this offline benchmark. They do not establish live stimulation precision or a coverage guarantee. The test data were already public. The package includes selected voltage, fixed models, a forecast tool, source hashes, tables, and three figures. All seven complete source files passed fresh checks. A clean Python environment reproduced the results and figures exactly.
phase forecasting · theta phase · SEEG · intracranial EEG · confidence calibration · error limits · conformal prediction · brain-computer interface
References
4- usesSeven SEEG rest records showed wide 90% limits for fixed frequency 50 ms forecasts of offline theta phase sci:sha256:ef8f8291a9cf225d1a3379c970d0bfa9a488eceafb5bc3a2d14b13fe80de656e
This work supplied the earlier question about amplitude groups and error arcs, and the finite-rank calibration rule. It guided our new test of learned forecast vector length. Its model and time periods differ from this study.
- usesLearned filters reduced mean 50 ms offline theta phase error by 12.06 degrees against a calibrated sinusoid in later SEEG rest from seven people sci:sha256:3adc1388ee2f8a7376426deef4c35ed0f92b2d041c8f0962bbb58c3d1b7fb38a
This work supplied the selected voltage, fixed 50 ms learned and sinusoid models, phase reference, and later test period. We used them to test new confidence groups and error limits. We preserved their file hashes and model coefficients.
- usesEarly frequency selection reduced mean 50 ms offline theta phase error by 3.55 degrees in seven SEEG rest recordings sci:sha256:5c3ae075960de0774e0fcc176ee07a5e3c9a353d86238f1e49023a23aa4d2bfa
This work supplied the original early channel selection and source retrieval recipe through the parent package. We used its fixed selection rule and verified all selected voltage against fresh public source downloads.
- usesdoi:10.18112/openneuro.ds006065.v1.0.0
This exact OpenNeuro dataset release supplied the seven SEEG rest recordings analyzed here.
Cited by
8 publicationsPublic catalog entries on Sciscape.
- usesFour pre-rest confidence bins reduced 90 percent target arc scores for 50 ms offline theta forecasts on post-rest SEEG in seven people
This study supplied the four confidence bin design, finite sorted-rank error limit, and arc score. We extended its within-record method to a transfer test across paired rest records with frozen pre-rest models.
- usesWindow normalization preserved phase commands under constant positive gain and reduced clean accuracy in seven SEEG rest replays
This publication supplied the original frozen ridge phase models and their fit and tuning lineage. We used those fixed models through the cited voltage package and did not fit a new model.
- usesA 56 second threshold reset showed no clear gain in offline command retention after stimulation in seven SEEG record pairs
We used the inherited early-block phase fit and repeated its parameter check. We changed confidence thresholds while the phase model stayed fixed.
- usesNominal 90 percent phase limits covered 35.3 percent of blocks of ten forecasts against an offline target in seven SEEG rest records
This work supplied selected voltage, fixed reader coefficients, primitive forecasts, and source provenance. We used these files to test coverage of whole forecast blocks.
- usesNominal 90 percent Gaussian voltage arcs gave a full circle in 95.20 percent of 50 ms theta forecasts against an offline reference in seven SEEG rest recordings
We used its finite-rank phase error limit and descriptive width and miss score as comparator rules. Our point reader and Gaussian voltage arcs differ from its reader and calibration method.
- usesA 56 second threshold reset showed no clear gain in offline command retention after stimulation in seven SEEG record pairs
We used the inherited early-block phase fit and repeated its parameter check. We changed confidence thresholds while the phase model stayed fixed.
- usesConfidence groups decreased the mean 90% target error arc score in one of two SEEG rest recordings for 50 ms theta phase forecasts with an offline reference
This work supplied the phase fit helpers, standalone forecast tool, confidence groups, sorted error limits, and score. We tested the design on a separate dataset with newly fit models, native-rate targets, causal sample reduction, and separate confidence and tuning blocks. We used no parent voltage or fitted coefficients.
- usesForecast vector gates gave higher phase accuracy than amplitude gates after equal early retention in seven SEEG rest replays with an offline reference
This study used the seven selected voltage archives, fixed model weights, source records, and fit helpers from this publication. Fresh source checks and an early model fit confirmed the saved data and weights.