A shared spectral profile did not improve SEEG word-recall decoding
In an offline analysis of 18 SEEG patients from OpenNeuro ds004789 (FR1, session 0), we compared a rank-1 decoder with a spectral profile learned from the other 17 patients and a patient-specific spatial map (S_trans) against an L2 decoder fitted to all channel-by-frequency weights (W_chan). S_trans had lower median test AUC than W_chan after 1, 2, 3, and 5 calibration lists (differences -0.010, -0.007, -0.009, and -0.023; Holm-corrected two-sided p = 0.69, 0.69, 0.18, and 0.042), and after 8 lists (-0.020; p = 0.003) and leave-one-list-out (-0.030; p = 0.002). The transferred profile was stable: its cosine similarity to the all-patient profile was at least 0.98 in all 18 leave-one-out profiles. It exceeded a profile estimated from each patient's own calibration data at 2 and 3 lists (+0.013 and +0.021; uncorrected p = 0.027 and 0.024), but that control was also not better than W_chan. These results show that the tested shared rank-1 spectral structure did not make calibration faster. The analysis was offline, used one session per patient, eight fixed frequencies, and one time window; higher-rank and region-specific structures were not tested.
SEEG · word recall · memory decoding · AUC · cross-patient transfer · spectral profile · calibration · brain-computer interface
References
3- supportsA cross-patient prior did not improve recall decoding after 1–3 calibration lists in 18 SEEG patients sci:sha256:c150a22bc4acaa236ebc6cbc6b644e8213d386d79c8fe35c7fff7ba324ff2691
This publication showed that adding a cross-patient decoder did not help after 1-3 lists. Our result agrees for a different transfer method (a shared spectral profile with a patient-specific spatial map).
- supportsA fixed cross-patient recall prior did not improve patient SEEG decoding with 1–3 labeled lists sci:sha256:f8f76c2bf2cb4118285e62c994861d8bf3a4ca90d3ea6e08b6dc69cf0933e2a6
This publication showed that a cross-patient model used as a fixed prior did not improve the patient decoder. Our result agrees: cross-patient spectral structure used as a hard constraint also did not improve few-list calibration.
- usesCross-patient SEEG recall decoding was above chance but weaker than a three-list within-patient decoder sci:sha256:d52ba56c23a69304d837e4ef5a9f3f20bcee59ebf2cd093faea38a0334a9fa4a
We used the archived per-word FR1 band-power features (features/sub-*.npz), the inclusion rule, the W_chan decoder settings (C = 7.2e-4) and the numpy logistic, AUC and Wilcoxon functions of this publication.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.