SEEG power reduced mean recall loss by 0.005 nats versus task features on excluded PAL2 lists in 13 people
We tested recall prediction on 554 unstimulated word pairs from 13 people in OpenNeuro ds005494 version 1.0.1. We used 62 remote bipolar SEEG channels. Each complete list was excluded from model fitting. At C = 0.01, adding 4 to 8 Hz and 70 to 110 Hz power reduced mean log loss from 0.4348 to 0.4297 nats. The mean gain was 0.0050 nats. The 95 percent person bootstrap interval was 0.0009 to 0.0098 nats. The exact sign-flip test gave p = 0.074. Gain over a model that used only the training recall rate was 0.0021 nats, with an interval of -0.0020 to 0.0064 nats. The group evidence is uncertain. Class-specific scores gave a mean gain of 0.0087 nats for pairs after a train and -0.0017 nats for lists with no train. This difference does not measure a stimulation effect. The package preserves native epochs, labels, features, folds, predictions, model parameters, figures, source checks, and reproduction code. This is an offline test of one small decoder, not proof of useful online memory control.
stereoelectroencephalography · SEEG · paired associate memory · recall prediction · brain-computer interface · logistic regression · OpenNeuro
References
4- usesAdding SEEG power increased mean test loss by 0.035 nats for recent stimulation prediction on matched PAL2 lists in 13 people at C = 0.1 sci:sha256:f88db094975ac78cd97c0af64026697b50e3836dfd1210e50180f5b7388fa057
Uses the prior study to define the open question and simple reference-model comparison. It shares the cohort but does not reuse its stimulation-history predictions.
- usesThe mean beta power time course contrast after stimulation was 0.072 dB in 13 PAL2 SEEG participants sci:sha256:28b3e1ff8da5625cfc749aa8d8c0d234884d66bc66fb01533c9abb496d1e0618
Uses native SEEG epochs, cohort selection, source metadata, pinned EDF source versions, extraction code, and timing audit. The new outcome is later paired-word recall.
- usesdoi:10.18112/openneuro.ds005494.v1.0.1
This is the exact OpenNeuro dataset version used for the archived PAL2 SEEG epochs and behavioral events.
- doi:10.1038/s41467-017-02753-0
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesA SEEG model fitted on lists with no train gave higher mean recall loss after stimulation than a rate model in 13 people at C = 0.01
We used the same native epochs, selected cohort, checked recall labels, feature and fit functions, source records, and extraction code. We fitted new models only on lists with no train and tested their predictions after a train.