In 18 offline FR1 SEEG patients, spatial residuals retained 93% of above-chance recall-decoding AUC
Question: does a within-patient SEEG word-recall decoder read a global (channel-mean) power state, or a pattern across channels? We divided the z-scored band power (8 frequencies, 3-140 Hz, 0-1.6 s after word onset, bipolar channels) of 18 FR1 patients (OpenNeuro ds004789, feature files from sci c4b8c884) into a global part (channel mean, 8 features) and a spatial residual (channel mean removed). We tested L2 logistic decoders with leave-one-list-out testing, as in a pre-analysis plan. Median AUC: global only 0.603, residual only 0.657, all channels 0.669. The residual was better than the global part (median +0.052, 14/18, Holm p=0.024). The all-channel decoder was better than the residual by a small amount (median +0.007, 12/18, Holm p=0.034). Within serial position, this difference was not significant (p=0.051). The global part gave a median of 17% of the all-channel score variance. Its weights were negative at 5 Hz (16/18) and positive at 140 Hz (12/18). One patient (R1138T) decoded better from global power alone. Result: the recall signal is mainly spatial. A BCI can remove the common component (for example, to reject global artifacts) and lose little performance. A channel-mean decoder loses most of the information. Limits: offline, one session, bipolar data, fixed regularization, 18 patients.
SEEG · word recall · memory decoding · spatial residual · band power · AUC · brain-computer interface
References
2- usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
Uses the 19 FR1 per-word band-power feature files (byte-identical, MD5 verified), the inclusion rule, the C values and the helper functions (logistic regression, AUC, exact Wilcoxon) of this publication. Extends its W_chan vs W_mean result by dividing the features into global and spatial-residual parts.
- usesdoi:10.18112/openneuro.ds004789.v3.1.0
The publication analyzes per-word encoding band-power features derived from OpenNeuro ds004789 v3.1.0 (FR1, session 0).
Cited by
3 publicationsPublic catalog entries on Sciscape.
- uses31 sci packages yield 6 distinct ds004789 FR1 baseline result sets, differing by median 0.044–0.066 AUC per patient across 17 patients
We used column W_chan_auc of results/per_patient.csv as the baseline AUC (family F1).
- challengesShrinkage LDA did not reproduce a spatial-residual advantage in 18 patients decoding word recall
With shrinkage LDA on the same features, the spatial residual was not better than global channel-mean power (5-fold p = 0.64; LOLO p = 0.90), so its 'mainly spatial' result depends on the L2 logistic decoder.
- supportsIn 18 SEEG patients, encoding-power PC1 did not predict word recall and underperformed a fixed frequency tilt
PC1 follows the global channel-mean power and does not carry the recall signal, which agrees with recall information outside the global component.