Offline within-patient SEEG decoding of verbal memory load (set sizes 8 vs 4) exceeded chance with two trials per class in 15 patients, with no significant advantage over cross-patient decoding
We measured how many labeled trials a within-patient SEEG memory-load decoder needs. Data: archived maintenance band power (5 bands, all bipolar channels) of OpenNeuro ds004752, 15 patients. Decoder: shrinkage LDA, set size 8 vs 4, 2 to 10 random training trials per class, 200 draws, test on all other trials. Features were z-scored with the unlabeled set size 6 trials. With 2 trials per class, the decoder with all channels x 5 bands gave mean AUC 0.62 (11 of 15 patients above 0.5, exact sign-flip p = 0.001). With 6 trials per class, it gave 0.68 (p = 0.001). A decoder trained only on the 14 other patients (channel-mean band power, 0 own labels) gave 0.58 (p = 0.06). The own decoder was better in the mean at all sizes (+0.04 to +0.10), but not significantly after Holm correction. Thus our prediction that the other-patient decoder is better for small calibration sets was not correct. To add other-patient trials to own trials did not help. A 70-150 Hz-only decoder was not better than the 5-band decoder. Limits: offline; random (not first) trials; unlabeled same-session normalization; small trial numbers.
SEEG · working memory · memory load · brain-computer interface · calibration · LDA · band power · AUC
References
4- usesCross-patient SEEG decoding of working-memory load was weak and did not reduce calibration trials sci:sha256:ef6e40afcc4e4a8257d2b53b2234a08c65c55bea478965440bbec82518399314
This publication reports the cross-patient load decoder. We compare it with a within-patient decoder at each calibration size.
- usesIn 15 SEEG patients (OpenNeuro ds004752), a band-power decoder trained only on memory loads 4 and 8 puts load 6 at the midpoint, but its output does not predict response time within a load sci:sha256:e2745563d558339564cbc0b329f8efc977263b7a2e3956a01d641ce57864f0b3
We use the archived ds004752 maintenance band-power features and BIDS sidecars of this publication.
- usesIn session 1 of 15 patients, IEDs did not decode memory load; masking raised SEEG AUC from 0.70 to 0.72 sci:sha256:632c406c34e1530eea3bb23516da5054eb6c64dbaa63783d6d4628a34fe21bbb
This publication computed the band-power features from the raw ds004752 SEEG data.
- doi:10.7554/eLife.78677
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesArchived SEEG pipelines agree on patient decodability rankings, but shared trial noise explains about 38% of patient variance
Per-patient load 8 vs 4 AUC table (column -1) is one pipeline in the cross-pipeline matrix.
- usesSet-size-6 trials did not improve a 4-vs-8 SEEG working-memory load decoder (ds004752, 15 patients, one session, offline band-power shrinkage LDA)
This publication computed the band-power features (origin of the input data).
- usesIn SEEG load decoding, noise correlation fell with distance but distance did not predict within-patient pair gain
This is the original source of the features and of the set size 8 vs 4 decoder setup.
- usesIn one session from 15 SEEG patients, averaging three same-load trials raised decoder AUC from 0.690 to 0.751, consistent with independent errors
Uses the archived ds004752 maintenance-window band-power features and provenance from this publication without change.