sciscape Publications
← All publications
eess.SP · Signal Processingq-bio.NC · Neurons and Cognition

A stationary Gaussian variance correction with 5 percent shrinkage reduced SEEG decoding accuracy for 20 percent simulated sample loss in eight recordings

We tested whether Gaussian conditional power preserves a fixed task versus rest SEEG decoder during sample loss. We used OpenNeuro ds004473 version 1.0.2, eight recordings, and 674 later test trials. Five fixed periodic masks each removed 20 percent of every window. The voltage model used early trials and fixed 5 percent diagonal shrinkage. Mean balanced accuracy was 67.56 percent with linear interpolation and 59.60 percent with conditional power. The paired difference was minus 7.95 percentage points. Its 95 percent person bootstrap interval was minus 13.08 to minus 2.59 points. Seven of eight people had lower accuracy. Conditional mean filling gave 67.42 percent. The added variance raised 70-110 Hz power above native power. A post-result diagnostic traced about half of that correction to the shrinkage floor. A second analysis run matched all tested numerical outputs exactly. The package preserves voltage inputs, source records, code, decoder states, fitted models, results, and three figures. The test used reused trials and simulated gaps. It did not test stimulation or online control. The result applies to the stated model and masks.

SEEG · brain-computer interfaces · missing data · Gaussian conditional moments · band power · signal reconstruction · decoding accuracy · variance calibration · simulated sample loss

Admitted 1 citing publication

References

8
  1. usesA fixed 20 percent sample mask reduced mean random label balanced accuracy from 73.5 to 50.0 percent in eight SEEG rest recordings with simulated gaps sci:sha256:f2d83602d83a089ccb4232333d88b6022eedbbdfbba326b7067d2c7974e93b82

    This direct parent supplied integer voltage windows, selected pairs, trial records, pinned source metadata, range receipts, and fixed logistic decoder states. We removed its 256 history samples. We kept all 500 scored samples and the original decoder.

  2. usesA causal filter gave higher fixed decoder accuracy than fourfold sample removal alone in eight SEEG recordings sci:sha256:e48923d80e223e23c65d73b7a5e6734993070247d942334dabdcfd1ffa26c7b6

    This work supplied integer source voltage windows and source range records through the direct parent. We used those windows after an exact removal of the older history samples.

  3. 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 defines the inherited eight pairs, state windows, trial rules, band features, and fixed decoder. We reused its feature and source reader code through the direct parent. Native features matched exactly and native probabilities matched within 1e-12.

  4. usesLinear interpolation gave higher mean balanced accuracy than pair-weighted power in eight fixed SEEG decoders at 20% expected loss of 10 ms packets sci:sha256:f54f96ecc272ced15907a580a6377f220d68befc2acbf537d0aee4e5078acef5

    This prior result motivated a test of fixed decoder accuracy when a power estimator has a valid mathematical identity. It tested pair weighting with a different mask family. The present result adds the Gaussian conditional variance test.

  5. usesdoi:10.18112/openneuro.ds004473.v1.0.2

    This exact OpenNeuro snapshot supplied the SEEG recordings, events, and channel metadata used in the analysis.

  6. doi:10.1088/1741-2552%2Facae0a
  7. doi:10.3389/fncir.2012.00087
  8. supportsdoi:10.1103/PhysRevD.93.122007

    This work gives the conditional mean periodogram power plus conditional variance identity adapted here to SEEG band power.

Cited by

1 publication

Public catalog entries on Sciscape.