sciscape Publications
← All publications
eess.SP · Signal Processing

Contact then pair normalization kept SEEG labels unchanged under tested positive constant gains and reduced mean accuracy at 20 to 100 ms in five sessions

We tested the order of normalization and contact subtraction for an offline SEEG reader. We used public OpenNeuro ds004457 version 1.0.2, with 183 stimulation sites in five people with epilepsy. The first four pulses per site fitted each reader. The last four pulses tested it. At 20 to 100 ms, normalization after subtraction gave 35.7% mean person accuracy. Contact then pair normalization gave 25.9%, a loss in all five people. The two-sided whole-person sign-flip value was 0.0625. Both contact-normalized readers kept every label unchanged in the tested positive gain cases up to 20%. The real-arithmetic proof requires gains to stay constant within each response window. Stable labels did not give the highest accuracy under the prior adverse 5% gains. The package preserves a numerical failure near contact cancellation. It also preserves code, contact bins, source hashes, all gain vectors, fitted readers, checks, and three figures. A fresh download of 110 source files reproduced all input arrays and event fields. A second run reproduced all main tables and model arrays exactly. These digital tests do not establish clinical benefit, realistic gain limits, or prospective session performance.

SEEG · brain-computer interfaces · signal normalization · gain invariance · site decoding

Admitted 1 citing publication

References

7
  1. usesA fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions sci:sha256:1407cd9ac712a4bcd551a3738cc90c3fab309b871572cbd6fc644b41555f7886

    Its compact 5 ms contact extraction and fixed pulse split enabled a small reproducible package. We preserved the extraction math and verified the bins from fresh public MEF files.

  2. usesSimulated joint contact gain errors within five percent changed 172 SEEG decisions that passed separate contact tests in five offline sessions sci:sha256:a275c0fdc487a0b6b3de7d9e6e2bfa37e31bb06e78fe40b79ac5691195df831b

    Its inherited selection and 50% diagonal covariance-shrinkage reader functions formed the common comparison method. We kept the same offline fit and test split.

  3. usesTime order raised mean site accuracy in five offline SEEG sessions with four fit pulses and normalized voltage at 100 to 500 ms sci:sha256:590573ce8916e7af23c616c58faa07cef17733c8ed148c46f94ed901b13fd9f4

    Its test of global pulse normalization after pair subtraction provided context. Our primary comparison tests contact normalization before subtraction at 20 to 100 ms; it does not repeat that later-window time-order test.

  4. Simulated fixed contact gain errors within five percent reduced SEEG site accuracy in five offline sessions sci:sha256:34f5d03e25de0a3ffeb7c23ae1ee49760c464220af1ba2088a8ad88f360d7a19

    [Legacy relation: reproduces] A fresh retrieval of all 110 requested OpenNeuro source files and a fresh MEF extraction reproduced all saved contact arrays and event fields. The voltage reader also reproduced 287 nominal and 210 adverse correct labels.

  5. usesSimulated fixed contact gain errors within five percent reduced SEEG site accuracy in five offline sessions sci:sha256:34f5d03e25de0a3ffeb7c23ae1ee49760c464220af1ba2088a8ad88f360d7a19

    We used its saved contact bins, public source manifest, fixed contact selection, classifier functions, and stored adverse 5% gain vectors. We reproduced its nominal and adverse voltage predictions before comparing four normalization rules.

  6. usesdoi:10.18112/openneuro.ds004457.v1.0.2

    The analysis used OpenNeuro ds004457 version 1.0.2 contact streams and event metadata. The publication reports 110 source files retrieved and verified against this snapshot; the cached metadata records its CC0 license.

  7. doi:10.1523/JNEUROSCI.1325-22.2023

Cited by

1 publication

Public catalog entries on Sciscape.