Covariance replacement gave no clear task accuracy gain after simulated loss of one known SEEG shaft in eight sessions
We tested whether early feature covariance can recover task decisions after the loss of one known SEEG electrode shaft. We used public OpenNeuro ds004473 version 1.0.2 and eight patient sessions. All fits used early trials. The test used 674 later trials and 35 simulated shaft loss cases. The intact decoder classified rest and windows before a key response. Covariance replacement kept this decoder and estimated missing power features from the surviving features. Its mean balanced accuracy was 69.13%, compared with 68.97% for training mean replacement. The patient mean gain was 0.16 percentage points, with a 95% patient bootstrap interval from -0.24 to +0.57 points. This test did not establish an accuracy gain. Mean squared error from the intact decoder score fell by 5.7% on average. This separate result did not ensure better task decisions. The test used a reused later-trial set, an exact loss flag, and retrospective quality selection. The package contains code, power arrays, source range recipes, all decisions, fitted models, and four figures. Fresh source retrieval reproduced all eight feature arrays exactly.
stereoelectroencephalography · brain-computer interface · shaft loss · covariance replacement · signal reconstruction · balanced accuracy · neural decoding · intracranial EEG
References
6- usesStored recovery models gave no clear mean accuracy gain after simulated loss of a known shaft in eight SEEG recordings sci:sha256:d37d0f9629d7ee35f30f183b444f728ef6d2bef470c6da8174b96d29fc9b90d4
We used its eight feature arrays, pinned source files, shaft mapping, reduced decoder reference, and source rebuild code. We reproduced the reference models and added covariance replacement and separate score error checks.
- usesLoss training raised mean accuracy by 0.34 points under simulated contact loss in eight SEEG decoders sci:sha256:884a92921286ae95392544128af6a4f343937c3b760e03bce164eec4599853d4
The earlier contact loss comparison helped define the new physical shaft loss question. The new test uses the whole shaft mask from the later stored model study.
- 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 work supplied the original eight-pair task decoder, power feature definitions, trial selection, and later-trial split. We kept those choices for the new covariance replacement test.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
The analysis used eight sessions from OpenNeuro ds004473 version 1.0.2.
- doi:10.1088/1741-2552%2Facae0a
- doi:10.3389/fnins.2022.858377
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.