Selection for surviving task utility gave no clear gain in lowest later accuracy after known simulated shaft loss in eight SEEG recordings at 16 contacts
We tested input selection under a known simulated SEEG shaft loss. Eight people from OpenNeuro ds004473 version 1.0.2 supplied 1,332 early fit trials and 674 later test trials. Each rule used eight bipolar pairs with no shared contact, for 16 contacts and 24 power features. The new rule maximized early task utility that survived the largest shaft loss. It increased mean lowest later balanced accuracy from 64.71% to 65.43%. The paired gain was 0.72 percentage points, with a descriptive 95% person bootstrap interval of -0.13 to +1.57 points and a two-sided sign-flip reference of p=0.15625. The utility objective improved, but the later accuracy gain was not clear. Fixed regularization checks also showed no clear gain. The package preserves frozen features, source metadata, raw audit bytes, pinned retrieval recipes, an integer selector, fitted readers, all fault predictions, three figures, and repeat checks. This is an exploratory offline task-state test in a reused cohort. It does not establish live BCI performance or stimulation benefit.
stereoelectroencephalography · brain-computer interfaces · fault-tolerant decoding · input selection · balanced accuracy · task-state decoding
References
6- usesBeta selection gave lower high-gamma reader accuracy than gamma selection in later trials from eight SEEG recordings sci:sha256:106f8a5c449affdc6d525f9a822fc65b3c0cda53615f1491b07c92a0b4ea78b1
This publication supplied the frozen three-band trial features, original source metadata, raw audit ranges, and extraction code copied with verified file hashes.
- usesAn eight-pair rule increased mean surviving high-gamma power after the worst simulated shaft loss in 36 SEEG test records sci:sha256:b5d10a3b7bd8a690c9522d78196853552e773420d35d2fb57e75374318bfbc13
Its summed-power objective after the largest shaft loss supplied the selection idea. This study tested a task-effect surrogate with distinct contacts and measured later decoder accuracy. It did not reuse the interictal data or decoder evidence.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
This publication supplied the original voltage and synchronized-trial extraction method and provenance used by the frozen feature source.
- One pair per shaft showed no clear gain in lowest balanced accuracy after a known simulated shaft loss in eight later-trial SEEG tests at 16 contacts sci:sha256:5f8c5a234d2daf29596b053b0c7a1a1f7aac5fdbdaf43faeac64da4b037739ad
[Legacy relation: reproduces] The total-utility and one-pair-per-shaft panels and later intact and lowest fault scores were reproduced in all eight people. Its independent path selector checked the two control rules.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analyzed feature and trial source is OpenNeuro ds004473 version 1.0.2; its saved dataset metadata records this exact DatasetDOI and CC0 license.
- doi:10.1088/1741-2552%2Facae0a
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesSelection for retained template energy gave no clear 100 to 500 ms AUC gain after known simulated shaft loss in seven SEEG sessions at eight contacts
Used its grouped-loss selection design as method context. This study tests template energy and later pulse-window AUC in a different SEEG dataset at eight contacts.