A gate with five thresholds gave outputs in one of eight offline SEEG records at a 20 percent error target
We tested how an error bound changed output from an offline SEEG reader. We used 2,014 trials from eight people in OpenNeuro ds004473 version 1.0.2. We kept fit, calibration, and later test trials separate. Each trial supplied one rest window and one window before a response. We checked five fixed confidence thresholds. Each gate had to meet a 20 percent error target in both true task classes. The adjusted binomial rule gave 165 outputs from 1,348 test windows. Ten outputs were wrong. Only one person had output. The rule based on calibration error alone gave 469 outputs, with 63 wrong outputs, across five people. Mean correct output fell from 27.37 to 10.53 percent of all later windows. At the 10 percent target, the adjusted rule gave no output. This rule needs at least 51 accepted calibration samples per class with no errors. Each record had only 19 to 47 calibration trials per class. We provide code, source checks, measured features, output tables, and three figures. The bounds require independent trials from one stable distribution. Task conditions changed across the ordered groups. These results do not establish live BCI performance or stimulation effects.
stereoelectroencephalography · brain-computer interface · selective classification · error control · Clopper-Pearson bound · calibration · offline evaluation
References
6- usesSelection for key identity had lower mean later key error than activity selection in eight offline SEEG records with tuned readers sci:sha256:8f807b882f06ccced0e3c3ddf31d99bf21b9b7ca83164353fda7c8a2ab07e7f8
We used its exact all-pair feature arrays, source metadata, pinned EDF range records, extraction code, and separate EDF audit functions. We fitted new readers and tested class error gates on separate calibration and later trial groups.
- usesSEEG key decoding from the first fast block had a mean AUC of 0.54 in later slow blocks in six patients sci:sha256:6407f4f45a2db401fef737a5f946c2656d5af25ba66dbcab30e4417db2c4fc72
The report identified a previously reported source key-timing concern for sub-11. That context supported the fixed omission check. We did not use its fitted readers or output scores.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
Its original measured features, retained-trial rules, and bounded source extraction recipe reached this study through publication 80063b09-a65a-4296-921c-07de4bab14ca. We preserve that source lineage, code, and raw range hashes.
- usesOffline SEEG label sets had 20.2 percent mean single-label error at a nominal 90 percent coverage target in eight recordings sci:sha256:e76d247dd4bb100a7cd2d2a24e80039a05928e303c3b41567c618d9770022089
Its distinction between label coverage and error among single-label outputs motivated the direct accepted-error target. We used its fixed fit/calibration/test groups and reader settings for a matched reference. We reproduced the 115 wrong single-label outputs out of 678 on the reused cohort.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
Source SEEG task data and metadata used in this offline analysis.
- usesdoi:10.1088/1741-2552%2Facae0a
The OpenNeuro dataset README identifies this as the accompanying study describing the source SEEG recordings.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.