SEEG readers with eight channels had mean choice AUCs of 0.491 at 50 ms lead and 0.495 at 250 ms lead within stimulus groups in 10 MetaRDK records
We tested eight-channel choice readers in public MetaRDK SEEG data. Two 400 ms signal windows ended 50 ms and 250 ms before movement. Readers trained on early trials and tested on later trials. Choice labels came from stimulus direction and task accuracy. We measured choice AUC within each stimulus direction, then gave equal weight to patients. We acquired 11 records and scored 10. The mean AUC was 0.491 at 50 ms lead and 0.495 at 250 ms lead. The paired difference was -0.004 (descriptive 95% patient interval -0.055 to 0.048; Holm p=1.000). The 250 ms AUC minus chance was -0.005 (Holm p=1.000). Under a centered 205-sample smoothing assumption, the 50 ms window can use input up to 149 ms after movement. The 250 ms window avoids this overlap from that filter. Source filter alignment was not verified. Earlier processing was not proved causal. The package contains compact data, source hashes, code, models, trial splits, checks, and figures.
stereo-electroencephalography · choice decoding · signal processing · filter support · brain-computer interface · area under the curve
References
4- usesSEEG readers fitted after movement lost 0.085 mean choice AUC before movement in 14 offline tests with eight channels at penalty 10 sci:sha256:afc193e3e7276ac2050c6767972df7be341ef82952059b93d7719fc61d52d876
We used the before and after movement question and the choice AUC definition within stimulus direction. This study compares two pre-movement lead times and adds a filter support calculation under a stated assumption.
- usesOffline SEEG readers trained on correct trials gave 48.6 percent balanced accuracy for inferred responses before movement on later errors in 15 MetaRDK records sci:sha256:42f128d29bab88ccc71b9ca6da8c9cdc803ac61baa43a6661cf0621e1c7bd89c
We used the trial masks, task metadata, pinned signal manifest, range retrieval design, and numerical core. Fresh full-source reads reproduced the three shared windows. We added the 250 ms lead window and trained on correct and error trials.
- usesdoi:10.18112/openneuro.ds006253.v1.0.3
This is the exact MetaRDK OpenNeuro release used for the analyzed SEEG derivative; its pinned commit and source hashes are preserved in the package.
- doi:10.1038/s41467-025-61744-8
The source article is cited for the task and signal-processing context, including its reported 400 ms Savitzky-Golay filter. The article and README differ on band counts, and the exact source filter alignment was not recovered.
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesEight added SEEG channels increased mean offline movement forecast error at penalty 10 in 10 MetaRDK records
This study used the pinned source manifest, eleven-record acquisition rule, source metadata, restricted retrieval code, and numerical solver from this reader study. We extracted new stimulus-aligned features. We did not use its choice forecasts or trial masks.