Adding 70 to 110 Hz power reduced pre-key pair loss across later task blocks in eight SEEG patients
We tested whether 70 to 110 Hz power adds information to 13 to 30 Hz power across task blocks. Training used earlier trials with one source task label. Testing used later trials with the other label. The test used eight SEEG patients, 11 eligible directions and 975 test window pairs from OpenNeuro ds004473 v1.0.1. Penalties and scales used training trials only. Adding high gamma reduced equal-patient mean pair loss by 0.0518 nats. The empirical 95% patient bootstrap interval was [0.0214, 0.0908]. Seven patient means improved. The exact two-sided sign-flip value was 0.015625, under independent patients and joint null sign symmetry. A check with no shared contacts gave a smaller gain and an interval that included zero. The task labels occur in blocks. Condition and time cannot be separated. The result applies to known window pairs in reused clinical recordings. The package preserves source features, pinned raw retrieval, code, all fits and scores, the local plan, checks and instructions. All source features were rebuilt from public signal ranges. An independent optimizer checked the primary fits.
stereo-EEG · high gamma · beta power · ridge regression · task transfer · patient bootstrap · pair classification
References
4- usesHigh gamma differences within shafts lowered mean paired window loss on later trials in eight SEEG recordings sci:sha256:40c59303ab98145d0c5ef87f5c0652bcc15966fc0ed2d42e59433eb3f1654969
We used its forward training-only penalty selection design. Each current source-condition model selects its penalty in three earlier forward folds with a 30-second gap.
- usesAdding high gamma lowered loss for pre-key window pairs in fixed tests of later trials from eight SEEG recordings sci:sha256:d2b1d5b8e9abf4ca0a8ac1d256a66ae506afe7db5b7c4371c851c7a1ed6d44b0
We used its half-second band features, trial metadata, raw record fixtures and pinned source retrieval code. We rebuilt all features from fresh public byte ranges. The new models learn one earlier task label and test the other later label.
- usesdoi:10.18112/openneuro.ds004473.v1.0.1
The analyzed SEEG recordings and metadata are from OpenNeuro ds004473 version 1.0.1.
- doi:10.1088/1741-2552%2Facae0a
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesPower ratio models had higher mean loss across later task labels in eight SEEG patients at a fixed penalty
This work uses its chronological task label transfer design, minimum trial counts, and time gap. Our model predicts each window and tests the ratio restriction.