Re-centering preserved SEEG detector accuracy under simulated gain drift, but its advantage over shape features was not significant
In archived SEEG key-press features from 8 subjects (OpenNeuro ds004473, task SlowFast), label-free causal re-centering of HG + beta features was compared with spectral-shape features under simulated per-channel gain drift. A shrinkage LDA was calibrated on trials 1–100 and tested on trials 101 onward at a fixed threshold; drift offsets of 0, 1, 2, and 3 dB used 200 draws per subject and level, with the same seed-0 draws as the earlier study. Re-centering used both windows from the previous 20 test trials and no labels. At 3 dB, mean balanced accuracy was 0.754 for re-centred HG + beta, 0.713 for shape, and 0.636 for unadapted HG + beta. The re-centred minus shape difference was +0.041 (7/8 subjects positive; exact two-sided p = 0.039, Holm p = 0.070), so it did not pass the pre-registered Holm correction. At 0 dB, re-centering differed from unadapted HG + beta by -0.003 BA (non-inferiority p = 0.035, Holm p = 0.070); non-inferiority was not shown after correction. Under a 3 dB linear ramp, mean BA was 0.753 for re-centred HG + beta and 0.699 for unadapted HG + beta. Under a constant per-channel offset, re-centering removes the offset exactly after the first test trial by design. The findings are limited to simulated broadband drift, 8 subjects with one offline session each, and a task design with a known 50/50 rest/press mix; real frequency-dependent, correlated, or step-like drift was not tested.
SEEG · intracranial EEG · gain drift · causal re-centering · spectral-shape features · shrinkage LDA · balanced accuracy · simulated drift
References
5- usesBeta power increased later-trial SEEG key-press AUC in 6 of 8 subjects sci:sha256:b76113878b22b13b8fce6b4b57cf321bb22ce16b618b55ae02c2ef42b972f816
Origin of the ds004473 HG and beta feature files and lda_np.py.
- supportsGain-invariant spectral features did not improve cross-session SEEG accuracy, while session centring did sci:sha256:cec553f54e61d341282e13c0298644d6aa4b8c0f9c1cd2b44db16f7d64351e9e
Same direction in a different task: centring of full band-power features beat gain-invariant shape features (+0.041 BA, 7/8), but not significant after Holm correction here.
- usesShape features preserved fixed-threshold SEEG key-press accuracy under 3 dB gain drift but reduced AUC without drift in eight subjects sci:sha256:71cd929f12b8eb5295b74f6f43a8ae17b00b8e49e14c7f00ea756c4edcc598bf
Uses its archived features, LDA code, train/test split and the same seed-0 drift draws; tests the open gap in its limits (no comparison with label-free re-centering). Its FULL and SHAPE numbers are reproduced exactly.
- supportsCausal feature re-centering did not improve later-trial SEEG key-press detection AUC in 8 subjects sci:sha256:ef882f7641e97013d264081ec7074a3c27da90a7852725f482be65f41917ed95
Causal re-centering on the previous trials had a small no-drift cost here too (BA -0.003, AUC -0.0025 for K = 20).
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The archived HG and beta features used in this study are documented as originating from OpenNeuro ds004473 snapshot 1.0.2 (CC0), task SlowFast.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.