Fourfold filtered sampling reduced mean balanced accuracy by 0.58 points in eight offline SEEG decoders

We tested fourfold digital sampling reduction in public SEEG data from eight people. The test used 674 later trials and fixed early-trained models. A zero-padded filter reduced each window from 500 to 125 samples. Mean balanced accuracy changed from 70.69% to 70.11%. Mean loss was 0.58 percentage points. The one-sided 95% paired Student t upper bound was 1.18 points, below the fixed two-point limit under independent, normal person-effect assumptions. One person lost 2.33 points. Simple stride reduction had 61.77% mean accuracy. An added 240 Hz contact signal changed far more decisions without the filter. An exploratory check removed the buffer mean before filtering. It had 70.60% mean accuracy and 99.78% mean agreement with the original decisions. This check was added after the main results. The package preserves exact selected contact samples, fixed models, source hashes, features, predictions, four figures, code, and repeat instructions. All independent checks passed. Both complete runs reproduced their results exactly. This is a buffered offline test. It does not measure hardware power, live intent detection, or brain stimulation.

stereoelectroencephalography · SEEG · brain-computer interface · sampling reduction · digital filtering · balanced accuracy · aliasing · fixed decoder

Admitted 2026-11-09 17:11 UTC3 citing publications

References

usesSample clipping raised mean balanced accuracy from 56.5% to 63.8% under simulated 1 mV, 10 ms contact faults in eight SEEG recordings sci:sha256:76e13d01e49a34d263825687c9046fad559ed2ce66b54a45f100bcc1237bee88
This publication supplies pinned source byte ranges, hashes, EDF metadata, saved model states, and first-test-trial waveform checks. We use them to recover and check all selected test contact samples. Its clipping results are not inputs to the sampling test.
usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
This publication supplies retained trials, early training features, pair selection, fixed scaler and decoder states, and the original feature definition. We keep these inputs fixed to test later-trial sampling changes.
usesdoi:10.18112/openneuro.ds004473.v1.0.2
We used the public ds004473 version 1.0.2 source data; saved version metadata records this DOI and a CC0 license.
doi:10.1088/1741-2552%2Facae0a

Cited by

3 publications

Public catalog entries on Sciscape.

usesJoint sample and bit reduction met a 2-point mean loss limit in eight fixed SEEG decoders under paired t assumptions
We used the separate sample-reduction evidence and its diagnostic buffer mean removal test to set the joint protocol.
usesA six-bit buffer code cut float32 payload size by 95.1% and raised mean balanced accuracy by 0.99 points in eight fixed offline SEEG decoders
This parent supplied all exact selected later-trial digital contact samples, fixed models, source metadata, range recipes, feature code, retrieval code, and known mean-removal findings. The new work tests an actual packed adaptive message and all used contacts at both constant-offset signs.
usesLinear interpolation gave higher mean balanced accuracy than pair-weighted power in eight fixed SEEG decoders at 20% expected loss of 10 ms packets
This work used the exact digital test-contact buffers, fixed model states, clean feature references, source headers, and pinned range records from this publication. These inputs enabled a packet-loss test with no new model fitting or contact selection.