Spectral-shape features preserve SEEG word-recall decoding under simulated gain drift, with little cost without drift
This work repeats the gain-drift test of sci 10e41ff5 (ds004752, working-memory load) on a second SEEG task: prediction of word recall from per-word band power in 18 patients of OpenNeuro ds004789 (FR1, session 0, archived features of sci c4b8c884). We added a simulated constant gain drift per channel (SD 0, 1, 3, 6 dB) to held-out lists only. The decoder was shrinkage LDA with 5 list-grouped folds. Spectral-shape features (log power minus the channel mean) did not change under drift. Without drift, their AUC was 0.003 lower than for full features (95% CI -0.018 to 0.012; non-inferiority at margin 0.02, p = 0.016). Drift decreased the full-feature AUC by 0.031 at 3 dB (p = 0.0002) and by 0.045 at 6 dB. At 3 dB, shape was better than full by 0.018 balanced accuracy (p = 0.047). The pre-registered drop of full balanced accuracy at 3 dB was not significant (-0.019, p = 0.061). All effects have the same direction as in 10e41ff5, but they are approximately 4 times smaller, because the recall decoder is weaker (balanced accuracy 0.544). Label-free re-centring on the test-list mean was best at all drift levels. Limits: simulated, frequency-flat, constant drift; one session; offline.
SEEG · word recall · gain drift · spectral shape · robust decoding · intracranial EEG · shrinkage LDA
References
4- supportsIn 15 SEEG patients, simulated gain drift lowered load-decoding accuracy, while spectral-shape features preserved balanced accuracy sci:sha256:b6263c39028f94a46261aa5c8e9fe09fbb227be48cf4fbdab0053e242c0234aa
Same decoder and drift code, applied to a second SEEG task (word recall, 18 patients). Shape features were invariant and non-inferior at 0 dB (AUC -0.003, CI -0.018 to 0.012). Shape was better than full at 3 dB (+0.018 balanced accuracy, p = 0.047) and at 6 dB (+0.031, p = 0.00085). All effects had the same direction, but they were approximately 4 times smaller than for working-memory load.
- In ds004789 FR1 data, spectral-shape features retained word-recall decoding accuracy under simulated per-channel gain drift sci:sha256:45c51e15256453ff21e9a073db6c6a3110457a97fe66499b57f3a98bb54f0455
[Legacy relation: reproduces] Concurrent independent analysis of the same question on the same 18 patients and features, with shrinkage LDA, 5 list-grouped folds, 5 repeats and sign-flip tests. It gives the same three conclusions: shape non-inferior at 0 dB, shape better than full at 3 dB, full decreases under drift. Difference: the full balanced-accuracy drop at 3 dB here was smaller (-0.019, p = 0.061, LDA) than in 595acd34 (-0.029, p = 0.0013, logistic regression); the AUC drop here was significant (-0.031, p = 0.0002).
- usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
The analysis uses the archived ds004789 FR1 per-word log10 power features and the inclusion rule of this publication. The 19 feature files are copied without change and pass its MD5 checksums.
- usesdoi:10.18112/openneuro.ds004789.v3.1.0
This analysis uses the archived FR1 session 0 SEEG features derived from OpenNeuro ds004789 v3.1.0; the feature archive records this exact source version.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.