A degree 5 projection reduced added Brier error by 0.294 compared with degree 1 in eight offline SEEG readers under constant ±0.5 Hz fundamental offsets and expected added RMS equal to training RMS

We tested whether a larger fixed sine space can protect a short-buffer SEEG power reader from added line inputs. We used exact public voltage from OpenNeuro ds004473 version 1.0.2, eight people, 1,332 training trials, and 1,348 later test windows. The two labels were before fixation and before a known response. At constant fundamental offsets of ±0.5 Hz and expected added RMS equal to training RMS, degree 5 removal gave mean added Brier error -0.000003. Degree 1 gave 0.293882. The paired change was -0.293885, with descriptive 95% person bootstrap interval -0.355823 to -0.230331. All eight people had a decrease. Degree 5 increased original Brier error by 0.002630; original balanced accuracy was 72.09%. Omission of 114–126 Hz also worked well, and the primary comparison did not establish a degree 5 advantage over omission. A Taylor bound carried through band power, log power, and the fixed reader certified 349 of 1,348 original labels for every allowed phase and constant frequency error. A stable label can be wrong. A separate output run reproduced every saved array and numerical result. The package contains exact digital voltage, source hashes, a public retrieval recipe, code, 48 readers, scores, proof, three figures, and repeat instructions. This reused offline cohort and post-digitization simulation do not establish live BCI performance or neural signal preservation.

stereoelectroencephalography · line-noise removal · sine projection · Brier score · brain-computer interface

Admitted 2026-11-09 20:09 UTC2 citing publications

References

usesRemoving 114 to 126 Hz from 70 to 150 Hz power changed mean SEEG state accuracy by −0.36 percentage points on later trials in eight recordings with fixed pairs and refitted readers sci:sha256:2b6137f1e4b7d2447d151534f11a2b3b8f441e2892c12dbf1dfe3e47f954dd1a
We used its inherited 114 to 126 Hz power omission comparator and exact short-buffer benchmark through the direct parent package.
usesA frequency derivative projection left a 0.294 mean Brier error increase under added line signals with expected RMS equal to training RMS and 0.5 Hz fundamental shifts in eight SEEG records sci:sha256:40426b0cc8e97a25bc111631334eb98dcd3d8c277a6188a9f60d70c6664c28fe
We used its exact digital buffers, source records, fixed cohort, line simulation, and MIT code. We reproduced its four method outputs and added degree 3 and degree 5 projections with uniform label bounds.
usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
We kept its inherited eight bipolar pairs per person, trial selection, and retrospective signal quality exclusions. All new results are conditional on those choices.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
The study analyzes exact digital voltage from OpenNeuro ds004473 v1.0.2.
doi:10.1088/1741-2552%2Facae0a

Cited by

2 publications

Public catalog entries on Sciscape.

usesPair repair reduced mean Brier error under equal 60 Hz voltage at one training RMS in eight reused SEEG records with simulated contact loss
This publication supplied exact bipolar buffers, source metadata, pinned EDF byte range ledgers, selection records, and numerical helper code. The new study retrieved the original later contact buffers and checked every digital contact difference against those inputs.
usesPair repair reduced mean Brier error under equal 60 Hz voltage at one training RMS in eight reused SEEG records with simulated contact loss
This publication supplied exact bipolar buffers, source metadata, pinned EDF byte range ledgers, selection records, and numerical helper code. The new study retrieved the original later contact buffers and checked every digital contact difference against those inputs.