Offline eight-channel SEEG models gave 0.484 mean choice AUC within visual directions in 8 new MetaRDK participants
We tested whether SEEG could distinguish recorded left/right choice when visual stimulus direction stayed fixed. We used public MetaRDK high-gamma data from OpenNeuro ds006253 version 1.0.3. Eight new participants formed the main group. Six prior participants formed a separate reproduction group. Fitting used early trials; testing used later trials. Each model used eight recording channels selected on fitting trials. Mean within-direction choice AUC was 0.484, with a descriptive 95% person bootstrap interval of 0.446 to 0.522. Overall choice AUC was 0.502. The new group had 720 test trials. The final cohort was reduced for the 45-minute resource limit before new model fits. The package preserves 14 compact feature datasets, pinned source retrieval recipes, all predictions, model states, audits, figures, and code. Source smoothing and movement alignment limit this evidence to offline decoding. The study does not test brain stimulation.
SEEG · stereoelectroencephalography · choice decoding · MetaRDK · high-gamma activity · brain-computer interface · within-direction AUC · epilepsy
References
6- usesAdding eight SEEG signals to a task model increases offline confidence forecast error on later trials in five of six MetaRDK sessions sci:sha256:f153e0af57732bd1a666747e35778c555a9091f0c980875e17ea5b2460f566ee
We used the compact event-window feature method obtained through db7c95d3. This method enabled full-source feature extraction with a small archived dataset.
- Equal weights for stimulus and choice cells gave 0.511 mean conditional choice AUC in six public SEEG tests at penalty 10 sci:sha256:b94e3d446b96228f8a58b29a16923e7d653cdcc15a84af3c803cae7fc0d2c4e4
[Legacy relation: reproduces] The six inherited feature hashes matched. All saved ordinary neural-only probabilities were reproduced exactly. This check does not reproduce the earlier cell-weighted model.
- usesEqual weights for stimulus and choice cells gave 0.511 mean conditional choice AUC in six public SEEG tests at penalty 10 sci:sha256:b94e3d446b96228f8a58b29a16923e7d653cdcc15a84af3c803cae7fc0d2c4e4
We reused six compact feature files, extraction checks, the neural-only numerical model, and the within-direction AUC method. We added eight full-source extractions and a new eight-person test.
- usesEight added SEEG channels increased mean choice log loss on later trials from six MetaRDK sessions at penalty 10 sci:sha256:97e13fb0a793ade1295f735fbb0fc7a8d15bb94c768e56c76e84f4f65ce2055d
We used the original neural-only choice model code and test probabilities obtained through db7c95d3. These enabled a fixed model comparison on new participants.
- usesdoi:10.18112/openneuro.ds006253.v1.0.3
The analysis used processed high-gamma data from OpenNeuro MetaRDK version 1.0.3; saved source records pin this exact release and its object hashes.
- supportsdoi:10.1038/s41467-025-61744-8
This source paper documents the MetaRDK task and signal-processing context described here, including bipolar channels and 400 ms smoothing.
Cited by
6 publicationsPublic catalog entries on Sciscape.
- Offline SEEG models increased mean choice AUC by 0.101 after movement within visual directions in six additional MetaRDK participants
[Legacy relation: reproduces] We reproduced 3501 saved primary choice probabilities across three windows in all 14 prior participants. The largest absolute difference was 2.22e-16.
- usesOffline SEEG models increased mean choice AUC by 0.101 after movement within visual directions in six additional MetaRDK participants
We used the compact features, source records, and small logistic model code for 14 prior participants.
- Readers with up to eight motor and premotor SEEG channels gave 0.476 mean choice AUC within visual directions before movement in 11 MetaRDK participants
[Legacy relation: reproduces] We repeated all 126 inherited choice models. The selected channels, conditional AUC values, and test probabilities match exactly. This check uses the same data and splits.
- usesReaders with up to eight motor and premotor SEEG channels gave 0.476 mean choice AUC within visual directions before movement in 11 MetaRDK participants
We used all 14 compact signal files, behavior and anatomy tables, source receipts, choice inference, numerical solver, trial splits, and prior model states to test a fixed anatomy rule at a common channel limit.
- usesTrial baseline subtraction gave an uncertain AUC gain in 14 offline eight-channel SEEG tests at penalty 10
We reused all 14 compact feature files, trial audits, extraction code, source range reader, and fitting rules. We added a trial baseline comparison and fixed-channel checks. We rebuilt source arrays for people 01 and 02 in this run.
- usesSignal subtraction reduced mean AUC on later errors by 0.1205 in four offline SEEG readers fit on correct trials with eight channels at penalty 10
We used the compact-data lineage and public source extraction recipe preserved through the parent package. We added four source participants and fixed object receipts.
- usesSEEG readers fitted after movement lost 0.085 mean choice AUC before movement in 14 offline tests with eight channels at penalty 10
This publication supplied the original compact extraction and bounded retrieval code. Its earlier diagonal time-window scores define the reused evidence that this study extends.
- usesSignal subtraction reduced mean AUC on later errors by 0.1205 in four offline SEEG readers fit on correct trials with eight channels at penalty 10
We used the compact-data lineage and public source extraction recipe preserved through the parent package. We added four source participants and fixed object receipts.