Early cue training gave an uncertain mean gain in minimum AUC on 487 later SEEG trials in seven people

We tested whether early cue examples improve a fixed SEEG reader on later trials. We used public OpenNeuro ds004473 version 1.0.2 features from an archived sci study. Seven people supplied 404 eligible fit trials, 218 calibration trials, and 487 later test trials. Each person had separate rest and rest + cue readers. Both used the same fit trials, feature scale, and fixed ridge penalty. The primary measure was the smaller AUC for near response versus rest and near response versus cue. Its equal-person mean rose from 0.664 to 0.681. The mean change was +0.017, with a descriptive 95 percent person-bootstrap interval of -0.020 to +0.050 and a two-sided sign-flip reference p of 0.391. Five people had a gain and two had a loss. Three fixed sensitivity tests also had intervals across zero. Lower mean cue call rates came with fewer near-response calls. These results do not establish a mean device gain or equality. The package preserves data, 28 reader models, all scores, figures, code, source hashes, a raw range recipe, and independent checks. No continuous BCI or brain stimulation test was run.

stereoelectroencephalography · brain-computer interface · neural decoding · cue training · AUC · classification · epilepsy · signal features