Pair power ratios changed decoder decisions under simulated 5% contact gain errors in eight SEEG recordings
We tested small constant contact gain errors in public SEEG data from eight people. Early trials selected eight bipolar pairs and fit three task state decoders. Later trials tested gain 0.95 or 1.05 at one recorded contact before digital subtraction. Pair power ratios changed 1.75% of decisions as an equal person and case mean. Recorded balanced accuracy was 67.05%. Mean case accuracy was 67.26%. The primary recorded-minus-case difference was -0.21 percentage points. The descriptive 95% person bootstrap interval was -0.46 to 0.02 points. The exact two-sided sign-flip probability was 0.179688. The primary test did not show a clear mean accuracy change. Common shaft gains changed no ratio decisions. We preserve a 2.27 MB spectral dataset. It exactly reconstructs band power for constant gains at either contact under the stated estimator. We also preserve code, source hashes, raw audits, scores, and three figures. Direct waveform checks and a full numeric repeat passed. This offline artificial gain test does not establish natural hardware faults, online intent detection, or brain stimulation performance.
stereoelectroencephalography · SEEG · power ratios · contact gain · brain-computer interfaces · spectral power · logistic regression
References
5- usesBand power ratios retain 67.0% balanced accuracy under simulated constant voltage gains per shaft in eight SEEG recordings sci:sha256:fa4b2be456049aafc97572ea0a4d44239abb9b5dc00341d8bcb7a0fd368bf201
Used its pair and shaft power ratio transforms and common shaft gain result to define the boundary tested here. The new study applies small unequal gains to recorded contacts before digital bipolar subtraction and preserves exact spectral moments.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
Used its exact retained trial records, candidate power features, source metadata, pinned byte-range hashes, and EDF extraction helpers. Fresh retrieval matched every selected native feature. Selected pairs and clean absolute accuracy also matched.
- usesCovariance gates increased mean rejection of faulty and recorded inputs under simulated contact gain 4 in eight offline SEEG tests sci:sha256:f6f73c27ffdcc0cee8e6993116ed7fca70712e0061e5ed6955e88a43107e3d96
Read its larger single-contact gain fault and input rejection study to define prior scope. This study tests small contact errors in fixed ratio decoders. Its contact data and gate scores were not copied.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
Analyzed SEEG source data: OpenNeuro ds004473, version 1.0.2.
- doi:10.1088/1741-2552%2Facae0a
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesEndpoint tests missed seven decision changes between simulated contact gains 0.8 and 1.2 in eight SEEG records
We use its exact contact spectral moments, fixed pair power ratio models, chronological split, candidate features, raw audits, and source hashes. We extend selected gain tests to continuous gain bounds and interior failure witnesses.
- usesQuadratic bounds certified 39.4 percent of later SEEG decisions for simultaneous constant contact gain changes of up to five percent in eight recordings
We reused the compact bipolar/common/cross spectral moments, all candidate features, fixed split, source recipes, and parent models. We derived and tested bounds for simultaneous changes at all contacts.
- usesCommon power and band contrast had mean AUCs of 0.750 and 0.753 in 28 locally calibrated second SEEG records outside onset shafts
This earlier motor-task test identified a limit from different contact gains before bipolar subtraction. We use that scope limit when stating our gain assumptions. We read it after the primary result and used no data or fitted model from it.
- usesSimulated one-sample contact delays changed a mean 6.89% of later decisions in eight fixed bipolar SEEG readers
Used the single-contact gain audit extraction layout as a guide for physical contact cases, source range hashes, and later-trial raw audits. This study tests delay before bipolar subtraction.
- usesLinked band bounds proved more SEEG classes stable under 5% constant digital contact gain errors in eight records
We used its contact spectral moments, fixed pair ratio models, source features, splits, raw audit ranges, source reader, moment builder, and versioned retrieval records. These inputs enabled bounds for all simultaneous unequal contact gain changes.