Loss-specific fits did not establish a mean accuracy gain at threshold 0.5 under known single-contact loss in eight SEEG sessions
We tested models for known physical contact loss in eight public SEEG sessions. The test used 2,014 trials and 106 single-contact cases. All new fits used the original early labels. Mean later balanced accuracy at threshold 0.5 rose from 70.00% with mean replacement to 70.45% with a loss-specific fit. The primary exact patient test gave p = 0.0625. It did not establish a mean accuracy gain. A planned secondary test showed higher ROC AUC, from 0.7572 to 0.7627, with Holm-adjusted p = 0.015625. Accuracy against prior loss training was uncertain. The package includes a physical-contact benchmark, frozen model bank, all case predictions, features, figures, code, and source records. Raw audit checks passed for all eight patients. A full feature rebuild passed for one patient. These offline results assume perfect loss detection within one session. They do not test live control or stimulation.
stereoelectroencephalography · SEEG · brain-computer interface · contact loss · logistic regression · balanced accuracy · ROC AUC · decoder robustness
References
5- usesLoss training raised mean accuracy by 0.34 points under simulated contact loss in eight SEEG decoders sci:sha256:884a92921286ae95392544128af6a4f343937c3b760e03bce164eec4599853d4
We used the eight preserved feature files, chronological splits, physical contact masks, loss-training method, and source records. We reproduced its reference models and scores. We added separate early-data fits for each known loss.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
The inherited feature data and extraction code came from this publication. We retained its event windows, trial exclusions, pair selection rule, and training-only scales.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
This version of OpenNeuro ds004473 supplied the SEEG data and physical contact identities used in the analysis.
- supportsdoi:10.1088/1741-2552%2Facae0a
The source study documents the SEEG recordings and cued two-choice task used here.
- doi:10.1038/ncomms13749
Cited as background on decoder robustness under electrode loss and changing recording conditions.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesEarly loss tests did not show better contact priorities than coefficient size under simulated loss in eight SEEG recordings
This study uses its known single-contact loss setting. A missing contact disables every selected bipolar pair that uses it. The new question concerns prediction of later loss, with no model refit.
- usesEarly loss tests did not show better contact priorities than coefficient size under simulated loss in eight SEEG recordings
This study uses its known single-contact loss setting. A missing contact disables every selected bipolar pair that uses it. The new question concerns prediction of later loss, with no model refit.
- usesJoint calibration reduced rest windows with any call across intact and known simulated single contact loss cases in eight offline SEEG tests
This publication defined loss at a physical SEEG contact. We use that definition to remove every selected bipolar pair that shares the missing contact.