Causal 100-ms phase prediction in awake SEEG is near chance, although reliable-peak pairs remain better
We measured 100-ms phase prediction error using a causal one-sided Hann-windowed FIR estimator in 19 awake SEEG patients from OpenNeuro ds005624. We reused 2,203 bipolar pairs, peak frequencies and reliable-peak flags from an earlier analysis. On the held-out half of one 60-s segment per patient, pooled median causal error was 78.0° for reliable-peak pairs (n=958) and 85.6° for other pairs (n=1,245), near the 90° chance level. Reliable pairs had lower per-subject median error in all 19 patients (median difference -7.7°, exact two-sided sign test p=3.8×10⁻⁶). The median causal penalty relative to offline zero-phase prediction for reliable pairs was 45.8° across patients. A 3-cycle kernel reduced pooled errors to 69.6° and 81.2°. A synthetic check gave 0.0° error for a pure 8-Hz sine and 1.6° for a sine plus 1/f noise at SNR 1. These findings show that the reliable-peak ranking persists with this causal estimator, while its absolute error is near chance. This is an upper bound for one simple causal method; only one 60-s segment per patient was analyzed, the reference phase was estimated offline, and no stimulation was delivered.
SEEG · phase prediction · causal estimation · signal processing · reliable spectral peak · phase-locked stimulation · OpenNeuro ds005624
References
2- usesReliable 2–30 Hz peaks mark awake SEEG pairs with lower offline 100-ms phase prediction error sci:sha256:cae8fa429f99828f4267db98117493ea15a6bec414093f84309918f8a3dc4fd0
Pairs, peak frequencies, reliable-peak flags, offline phase filter and peaks.py code are used without change. This work measures the causal-estimator error that 9d9413e2 names as a limit.
- usesAfter lag calibration, phase at bipolar pairs two contacts apart differed by 63–71° in 21 awake SEEG patients sci:sha256:3028482333d1f5cb611a6ed9a504ef82edfce2f88f4abd5062bc4e166c858d73
fetch.py and the ds005624 S3 listing are used to fetch the same 60-s byte ranges again.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesIn 19 awake SEEG patients, filter choice caused most of the AR phase forecast error gap
Evaluation rules (calibration/test halves, bias fit, amplitude gate, 100-ms horizon) and the Hann 4.5 baseline.
- usesAutoregressive prediction reduced 100-ms phase error in sampled awake SEEG pairs, but its advantage depended on implementation
Uses its causal Hann-FIR estimator, scoring protocol (reference, bias calibration, gate) and code (causal_phase.py); its full-grid per-pair results are the baseline check. Tests the AR method that it named as untested.
- usesCausal AR prediction reduced 100-ms SEEG phase error, but median error remained 63° across 958 reliable pairs
Question, Hann FIR baseline code (causal_phase.py), segments, gate, calibration design and the 78-deg baseline that this work tries to improve. The FIR 4.5-cycle baseline was reproduced (78.1 deg).
- challengesIn 10 awake SEEG patients, ecHT and ridge Wiener reduce 100-ms phase error from 78 to 44–47 deg; AR remains near chance
Its near-chance causal error is specific to the long Hann FIR. ecHT and a ridge Wiener FIR give 44-47 deg on the same pairs and protocol.
- usesIn 10 awake SEEG patients, ecHT and ridge Wiener reduce 100-ms phase error from 78 to 44–47 deg; AR remains near chance
We use its pairs, code, reference filter and calibration/test protocol. Its Hann FIR 4.5-cycle error (78.0 deg) is reproduced here as 77.7 deg on 10 subjects.