Four bipolar SEEG pairs reach 76.4% mean accuracy for response and rest in later trials from eight patients at 250 ms after a key press
We tested whether a small SEEG decoder can work on later trials of the same session. We used 1765 trials from all eight patients in OpenNeuro ds004473, version 1.0.2. We selected four adjacent-contact bipolar pairs and trained a fixed linear model on the first 60% of eligible trials. We tested the last 40%. We used causal signal filters and a 500 ms power window that ended 250 ms after the key press. Mean accuracy across patients was 76.4% (95% patient bootstrap interval, 70.2 to 82.1%). All available pairs reached 76.8% mean accuracy. 8 of eight patient tests passed the block label-swap control after Holm correction. This is an offline test of task phases within one session per patient. It does not establish motor intent, finger identity, continuous online detection, or stimulation performance. The package contains the features, source metadata, exact raw-data retrieval recipe, all test predictions, code, and figures.
stereo-EEG · bipolar signals · brain-computer interface · neural decoding · temporal validation
References
2- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis used all eight patients from OpenNeuro ds004473 version 1.0.2; this exact-version dataset DOI is recorded in the source metadata.
- doi:10.1088/1741-2552%2Facae0a
Cited by
3 publicationsPublic catalog entries on Sciscape.
- uses250 and 500 ms SEEG windows gave 75.8% and 75.3% mean balanced accuracy on later trials in eight patients at known keypress times
This publication supplied the EDF reader, source metadata, fixed S3 versions, range hashes, and 500 ms reference features. We retrieved all 1765 ranges again and checked all 500 ms features exactly. We tested four averaging lengths at fixed event times in the same cohort.
- supportsEight bipolar signals selected on early trials gave higher mean key-press AUC than random sets on later trials in eight SEEG sessions
A separate early-to-later bipolar task-state test also gives useful later-trial discrimination with a small signal set. The present study adds nested parameter choice, 100 random signal sets per patient, and a rest-versus-rest control. The windows, source versions, trial eligibility, classifiers, and metrics differ. Both studies use the same underlying patient group, so this is not independent patient replication or a direct performance comparison.
- usesAt threshold 0.5, a fixed response-versus-rest SEEG decoder had higher mean activation before go cues than at rest in later trials from eight patients
We used the exact trial set, all-pair features, source metadata, pinned EDF range manifests, source extraction code, and fixed model settings. We retrieved all ranges again and added pre-go controls and a weighted task-negative model on the same later trials. This is a dependent extension in the same patient cohort.