An eight-pair SEEG readout reached 51.7% balanced accuracy across digits and words on later trials in 11 patients
We tested number magnitude across two visual forms in public SEEG data from OpenNeuro ds005415. Each model used eight bipolar pairs. All test trials followed all training trials. 11 patients met the fixed class-count rules. Mean transfer balanced accuracy was 51.7% (95% person bootstrap interval, 42.6 to 60.2%). The conditional permutation probability was 0.3611. Same-form accuracy was 52.1%. Before-cue transfer accuracy was 52.5%. The fixed test did not show transfer above its conditional label null. This result applies to the tested features, model, and within-recording split. It does not establish voluntary intent decoding or stimulation targets. The package includes source metadata, compact features, code, predictions, models, figures, raw audit segments, and hashed source retrieval recipes.
stereoelectroencephalography · number magnitude · cross-format decoding · brain-computer interface · balanced accuracy · conditional permutation
References
2- usesdoi:10.18112/openneuro.ds005415.v1.0.0
Source data: OpenNeuro Numbers dataset, version 1.0.0 (CC0).
- doi:10.1371/journal.pone.0313155
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesPower plus voltage gave 57.0% balanced accuracy for later format reading in 11 SEEG patients with eight pairs selected from early power
Used the clock match and fixed ridge reader design through the immediate feature source.
- usesA fixed SEEG reader showed no clear format gain after onset on unseen even numbers in 11 patients
This study uses fixed ridge-reader and timing methods inherited through the parent package. It changes the training weights, prediction targets, and identity split as specified in its saved local protocol.
- usesA window before late Cue contained samples after recorded onset in 187 of 226 beep trials from six public SEEG runs
Its description supplied checks for auxiliary sensors and shortened EDF contact labels before decoder scores. It also reported the short sub-80 source file. We confirmed that byte issue. We did not reuse its code, signal features, or readout results. Its test used visual number forms; this test uses beep trials.
- usesA fixed eight-pair SEEG reader reached 47.4% balanced accuracy between digits and spoken words on later trials in 11 patients
This study adapts the fixation clock match, fixed eight-pair ridge reader, common chronological split, and local fixed-model label null from this visual-form study. It replaces written words with spoken words and retrieves new spoken features. It reuses the patient cohort and some digit trials, so it is an exploratory extension rather than an independent replication.