Personal alpha spatial filters improved run 2 balanced accuracy over mean alpha power in 42 ds007987 adults after calibration within one visit

We tested one personal alpha spatial filter against mean alpha power from the same eight EEG channels. Each method used run 1 for its personal fit and threshold. Run 2 was the test. The source was OpenNeuro ds007987 version 1.1.0. The primary sample had 42 adults. Mean later-run balanced accuracy was 73.75% for the spatial filter. It was 68.17% for mean channel power. The paired difference was 5.58 percentage points. Its 95% person bootstrap interval was 2.96 to 8.27 points. The interval supports a mean accuracy gain for this fixed method in the tested sample. Both run pairs came from one visit. The package preserves all 172 source crop hashes, covariance data, code, a raw example, and reproduction instructions.

electroencephalography · EEG · alpha rhythm · spatial filtering · eye-state classification · balanced accuracy

Admitted 2026-11-09 18:10 UTC3 citing publications

References

usesClosed-eye 8–13 Hz log power was more repeatable than the closed/open ratio across two within-visit runs in 42 adults sci:sha256:6fe6528c7913bd33774ab44cf866aa4fca203ffe7eb0246608e81b7888ac0fbe
We adapted the header parser, source crop recipe, mean reference, channel rows, and quality rules from this work.
usesThe local 8–13 Hz eye-state contrast decreased on repeat during one visit in 42 ds007987 adults, with 60-second EEG and assumed channel row order sci:sha256:a35a11d99144f44958a3bacdf78755f42e316c068435e9ffbcfd87836b71c29d
We used the pinned source range ledger, source metadata, and prior quality data. The known cohort result informed the saved follow-up plan.
usesdoi:10.18112/openneuro.ds007987.v1.1.0
The analysis used OpenNeuro ds007987 version 1.1.0; all source crop hashes and the pinned source metadata are preserved.
doi:10.1016/j.neuroimage.2021.118809
doi:10.1049/iet-spr.2016.0520

Cited by

3 publications

Public catalog entries on Sciscape.

usesPersonal alpha filter gains over peer filters had a 95% interval that included zero in 42 ds007987 adults. Both methods used personal threshold calibration for later runs within one visit.
We used all 172 alpha covariance records, pinned metadata, and raw retrieval code from this archive. We used its personal filter rule for a new peer-filter comparison. The independent audit also reproduced its primary personal and mean-power accuracy and AUC values. The new question concerns the added value of a personal spatial fit after personal threshold calibration.
Personal alpha spatial filters showed no clear mean balanced accuracy gain over personal mean alpha power with an average reference one month after calibration in 60 ds004148 adults
[Legacy relation: reproduces] This publication supplied the fixed spatial filter, power baseline, and earlier within-visit result. We repeated the method in a separate cohort across a month. The original method code is preserved for credit and inspection.
usesA fixed nearest neighbor rule gave 71.0% mean accuracy for random labels in run pairs with shared EEG records in 42 ds007987 adults
Supplied all 172 alpha covariance inputs, raw example, headers, metadata, source range ledger, and extraction code. This study uses diagonal variances in a new fixed nearest neighbor test and checks four raw crops again.