Earlier amplitude calibration reduced later Brier loss for 50 ms theta forecasts against offline targets in 17 SEEG rest files from seven patients

We tested whether earlier forecast amplitude could give useful probabilities of later theta phase accuracy. The test used public OpenNeuro ds006065 version 1.0.0 and 17 rest files from seven patients. Accuracy meant a 50 ms forecast was within 45 degrees of a specified offline 4–10 Hz target. Each file used 75 seconds for a fixed logistic probability fit and a separate later 75-second test period. Patient mean Brier loss fell from 0.247548 for an earlier constant rate to 0.236642 for forecast amplitude. The mean gain was 0.010906, or 4.41 percent. All seven patients and all 17 files had positive gains. The descriptive patient bootstrap interval was 0.008471 to 0.013131. The exact patient sign-flip value was 0.015625, under independent patients and symmetric null gains. The study reuses known patients and periods. Offline target agreement does not establish biological phase truth or stimulation benefit. The package includes native inputs, source pins, code, all fitted probabilities, results, and reproduction checks.

theta phase · SEEG · phase forecasting · Brier loss · probability calibration · forecast amplitude · intracranial EEG · offline phase target · OpenNeuro