Angle agreement lowered mean Brier score for fixed 50 ms forecasts of offline theta phase in seven later SEEG rest records

We tested whether forecast angle agreement adds confidence information on a separate later rest record. We used public SEEG from seven people in OpenNeuro ds006065 version 1.0.0. Each person had four fixed contact pairs. Phase readers used 100 ms and 500 ms of past voltage. All reader and confidence weights came from earlier rest. The later test had 82,880 opportunities. Success meant that the short reader was within 45 degrees of a centered offline theta phase reference 50 ms after the cutoff. Adding angle agreement to both log output lengths reduced mean Brier score from 0.23017 to 0.22947. The reduction was 0.000700. The descriptive 95 percent person interval was 0.000201 to 0.001238. Six of seven people had a reduction. Average calibration bin error did not decrease. The strictest threshold slightly reduced successful yield. An additional quadratic-length comparison also gave a reduction. The package includes voltages, fixed weights, forecasts, figures, source checks, a stream reader, and reproduction code. Fresh full downloads matched all 14 source hashes and every retained voltage sample. This reused cohort and these future-dependent labels do not establish live timing or a stimulation benefit.

SEEG · theta phase · phase forecasting · confidence estimation · angle agreement · Brier score · intracranial EEG

Admitted 2026-11-09 19:18 UTC7 citing publications

References

usesAdding forecast angle agreement lowered mean Brier score for 50 ms offline theta phase labels in seven SEEG rest records sci:sha256:8290581fcefed7706cfd2662e71acb73cf04a00579455e9ed55dfecfd84fdd7e
This publication supplied the fixed phase readers, confidence models, and numerical functions. We tested these models on separate later rest records.
usesFour pre-rest confidence bins reduced 90 percent target arc scores for 50 ms offline theta forecasts on post-rest SEEG in seven people sci:sha256:15c4fe876db97ae39777d428f9e2bccefeccf9e0131b42f3d2d38ef99ccd7370
This publication supplied the paired rest voltages, pinned source metadata, and retrieval code. We checked every retained sample against fresh complete source files.
usesdoi:10.18112/openneuro.ds006065.v1.0.0
Source SEEG voltage and recording metadata for the seven people and 14 records analyzed.
supportsdoi:10.1038/s41467-025-59417-7
Source study for the stimulation and rest-recording context of the dataset; this publication scores rest voltage and does not estimate a stimulation effect.
doi:10.1523/ENEURO.0050-23.2023
doi:10.1016/j.neuroimage.2020.116761

Cited by

7 publications

Public catalog entries on Sciscape.

usesComplex voltage output length gave higher mean AUC than unit vector length for 50 ms SEEG phase success across files in 36 selected patients with one offline reference
Uses the earlier result as scientific context for forecast confidence and the distinction between output strength and success probability. The current test compares two fitting targets on a different dataset. No code or samples from the earlier confidence publication are reused.
usesStarting the timer at accepted crossings increased correct and incorrect output rates in seven SEEG rest replays with a 1 s holdoff
Use paired public SEEG voltage, fixed reader and confidence models, and source records. These files support the two timer tests.
usesWith baseline removal, separate readouts showed no clear 20–300 ms RMS excess over time controls after causal theta triggers in seven later intracranial rest records
We used its 14 exact paired-rest bipolar voltage files, pinned source metadata, fixed contact selection, and raw reconstruction code. We tested new coordinate-separated no-pulse endpoints. We did not use confidence scores or model forecasts.
usesA 30 s rolling threshold decreased local acceptance rate error in seven later SEEG-labelled rest records at a 25% target
This work uses its 14 selected voltage archives and seven fitted model archives. All inputs match their saved hashes. It also uses the saved pair selection and full source audit. The new test compares rolling and fixed thresholds on the same later records.
usesComplex voltage output length gave higher mean AUC than unit vector length for 50 ms SEEG phase success across files in 36 selected patients with one offline reference
Uses the earlier result as scientific context for forecast confidence and the distinction between output strength and success probability. The current test compares two fitting targets on a different dataset. No code or samples from the earlier confidence publication are reused.
usesSeparate confidence updates for four pulse phases increased mean Brier score in seven later SEEG rest records with a 1 s holdoff
This publication supplied the selected earlier and later voltage, fixed phase readers, confidence weights, and raw-source retrieval code. We kept these inputs fixed and tested new confidence intercepts at simulated pulse times.
usesFull confidence fitting lowered mean Brier score by 0.00016 against an offset update for joint 4–10 and 4–8 Hz phase success in seven later SEEG rest records
We used its 14 voltage archives, seven fixed model archives, numerical helper code, and pinned source retrieval code. We extended its single-reference confidence test to a joint phase event and compared an offset update with full fitting.