Fixed 1–80 Hz SEEG mode prediction gave a 3.0 percent mean error gain with an interval that included zero in 36 epilepsy patients

We tested a fixed spatial mode predictor in public SEEG data from 36 patients with epilepsy. The primary band was 1 to 80 Hz. We fitted covariance changes with the first 25 second block of each of two files. We tested the prediction with later blocks that used separate filter windows. The predictor used 10 percent of the modes, rounded up, and a weight of 0.5. Its equal-patient mean error gain against zero prediction was 3.04 percent. The 95 percent patient bootstrap interval was -1.67 to 7.00 percent. Error was lower in 26 of 36 patients. The interval included zero. This test did not establish a mean benefit for the primary predictor. All nine planned secondary settings are reported. The scores changed with prediction weight. The package supplies a benchmark that removes an overall power multiplier and gives the same answer in two contrast bases. It preserves inputs, fitted spectra, source versions and hashes, code, results, and reproduction instructions. The result concerns short estimated covariance differences in one epilepsy cohort. It does not identify neural sources or establish clinical use.

SEEG · intracranial EEG · covariance prediction · spatial modes · signal processing · epilepsy · prediction error · bootstrap

Admitted 2026-11-09 19:32 UTC0 citing publications

References

usesThe mean rank-one/full loss ratio was 1.1022 for 36 SEEG patients in separate file tests at 1–80 Hz and weight 1 with fixed training scales sci:sha256:4e654bc19e848662c50c8c7af8afd1b5ba7ed49b3029251f3e03dcd9d0a45d51
We used this covariance forecast study to state the scope of the new contribution. It tested pair-specific covariance compression. Our target is the signed normalized covariance difference between two files. Its numeric scores did not enter the new analysis.
usesBipolar 1–80 Hz SEEG correlation magnitude maps had positive correlations between file excerpts after signal, shaft pair, and log distance controls in 36 epilepsy patients sci:sha256:36170aff18fbecc7f57d9bf3c205be69a8ce1d45617c7e9600a74df38e629542
We used all 72 covariance files, the separate filter windows, and the raw EDF reconstruction code from this publication. These inputs enabled the split prediction test. The new package keeps the input hashes, source versions, and code notices.
usesWith scale matching and a 1% ridge, fixed 50% covariance pooling lowered 1–80 Hz SEEG prediction loss in 29 of 36 epilepsy patients sci:sha256:4e89c74307bf8f48b5bd0cda9e74c9554126ad2e463998a15777a88cab1ff3cf
We used its covariance pooling test as related prior work. It predicted a covariance matrix after a local block. Our test predicts the difference between two files. We do not claim a new independent cohort. Its numeric scores did not enter this analysis.
usesIn 36 patients, fixed 1–80 Hz SEEG contrasts showed larger covariance shape changes between files than within files sci:sha256:daf70bbdcc88f893fea25fa9870f5621a6fd8289ee4133a9ec263f83268cd815
Its covariance shape distance result supplied the question for this study. It measured the amount of change between files. We tested whether fitted change modes predicted a separate later covariance difference in the same cohort.
usesTwo bases of the same SEEG contrast space gave different 1–80 Hz correlation participation ratios in 36 patients sci:sha256:901bade82e1972a491fb8b27d87ddf5ae157f439c20fb612d6d24768419473f2
We used its fixed contact contrast space and contact geometry rule. These rules enabled the score checks in two bases of the same contrast space. We did not use its dimension score as the prediction target.
usesdoi:10.18112/openneuro.ds004100.v1.1.3
The analyzed covariance inputs and raw EDF excerpts come from OpenNeuro ds004100 version 1.1.3, as recorded in the publication and saved version-pinned metadata.
doi:10.1109/86.895946

Cited by

0 publications

Public catalog entries on Sciscape.

No incoming citations on Sciscape.