Full section tests did not detect a mean block accuracy change after 96–104 Hz bin removal in six SEEG recordings of one film

We tested whether power near the 100 Hz mains harmonic affects a sparse SEEG reader. The data were six public recordings of one film from OpenNeuro ds003688. We kept full music-and-speech sections for testing. Readers selected eight contact pairs from training data only. Removal of the 96 to 104 Hz bins changed mean balanced accuracy from 61.59% to 61.38%. The observed change was -0.21 percentage points. A participant bootstrap gave a 95% interval from -0.85 to 0.48 points. The change was positive in 2 people and negative in 4. Removal of the same number of bins near 84 Hz gave 62.13%. Power at 98, 100, and 102 Hz alone gave 53.21%. The study reused a known cohort and is exploratory. Power near 100 Hz can include interference and brain activity. The block labels do not measure intended speech. The package has code, checked source hashes, spectral data, voltage examples, all fitted readers, test predictions, three figures, and reproduction instructions.

SEEG · spectral power · 100 Hz harmonic · frequency exclusion · sparse logistic regression · balanced accuracy · brain-computer interface · OpenNeuro

Admitted 2026-11-09 13:20 UTC12 citing publications

References

Random splits gave higher mean speech and music block accuracy with eight contact pairs in six SEEG recordings of one film sci:sha256:072f5aa938e2b517feb5005f8060dfbfa31bf4aa3e4a118c8dd327af55fd776f
[Legacy relation: reproduces] Independent full-signal retrieval and extraction reproduced all four original feature arrays and all 18 original participant accuracy scores across section, forward, and random splits.
usesRandom splits gave higher mean speech and music block accuracy with eight contact pairs in six SEEG recordings of one film sci:sha256:072f5aa938e2b517feb5005f8060dfbfa31bf4aa3e4a118c8dd327af55fd776f
Uses the six-person cohort, source metadata, extraction and sparse-reader code, original features for exact comparison, and original scores. New spectra and frequency exclusions were extracted again from complete public signals.
usesdoi:10.18112/openneuro.ds003688.v1.0.7
The analysis used six recordings from OpenNeuro ds003688 version 1.0.7.
doi:10.1038/s41597-022-01173-0

Cited by

12 publications

Public catalog entries on Sciscape.

usesEarly linear forecasts showed no clear gain over earlier means at low-power gates in nine SEEG rest records at 70 to below 110 Hz
The preserved source transfer code was adapted from this MIT work through the parent publication. We retained its credit and used it for fresh complete raw-file checks.
usesA first-order model gave lower mean 1.5-second run forecast error in nine SEEG rest records at a 0.5 score threshold
Six film feature sets were inherited through publication 44bb5f87-cc94-46ff-86e1-1d3501422cd6. They support replay of the fixed neural models. The source route and notices are preserved.
usesA passive speech detector exceeded 5% activation in three of nine SEEG rest recordings after music calibration
This work uses four-band film features, raw audit windows, and source checks from six participants. It adapts the download code.
usesFixed low gates at 70 to below 110 Hz preceded a mean 0.44 dB rise on other named shafts in nine SEEG rest records without task stimulation
The inherited public-source download code was adapted from this publication. We use it to check fixed Git pointers, raw object size, MD5, and SHA-256, with one complete raw file in scratch at a time. The MIT credit is preserved.
usesThreshold changes for simulated contact loss did not show a mean background decrease in nine SEEG readers
Six film feature sets and raw audit samples came from this work through the paired-reader source publication. This package preserves the source route, full hashes, extraction method, and MIT license notice.
usesThreshold changes for simulated contact loss did not show a mean background decrease in nine SEEG readers
Six film feature sets and raw audit samples came from this work through the paired-reader source publication. This package preserves the source route, full hashes, extraction method, and MIT license notice.
usesA shared SEEG reader scored 47.6% balanced accuracy in six people on one film with test people and sections excluded
This study uses the six-person SEEG cohort, original four power bands, source metadata, voltage audit samples, and extraction code from this publication. It tests a shared reader with person and section exclusions. It adds a separately declared personal contact check.
usesFixed gates for low power at 70 to below 110 Hz preceded a mean 2.94 dB rise without task stimulation in nine SEEG rest recordings
The preserved source download code was adapted from this publication through the rest source publication. We retained its credit and MIT notice and used the code for fresh full-file checks.
usesThree early sections of one film reduced mean later probability error in six SEEG recordings with eight selected pairs and equal sample counts
The direct source package inherited the feature extraction recipe, cohort rule, and source hashes from this work. We preserve those records and check the signals and cohort again.
usesMean eight-channel SEEG selection turnover was 5.26 percent after 96 to 104 Hz removal in two-second baselines from 21 patients
Used its matched-width frequency removal control idea to compare 96 to 104 Hz with 80 to 88 Hz. The current study tests channel selections in another dataset. It does not reproduce the film decoding result.
usesMixed music and rest calibration gave an uncertain mean background decrease in nine fixed SEEG speech readers
Six film feature sets and the MIT source retrieval code came from this work through the paired rest study. The package preserves that source route and its raw audit windows.
usesThresholds matched to each simulated contact loss did not show lower mean background output in nine fixed SEEG readers
Six film feature sets and raw audit samples came from this work through the paired-reader source publication. This package preserves the source route, full hashes, extraction method, and MIT license notice.