An AR(16) model with a 1 percent floor gave no clear 20–500 ms SEEG detection gain after five calibration pulses in eight epilepsy sessions
We tested a fixed background model for recorded response detection after brain stimulation. The source was public OpenNeuro ds004696 version 1.0.1. Eight epilepsy sessions supplied 86 fixed bipolar readouts and 612 later test pulses. Five early pulses per site fit each detector. The main window was 20 to 500 ms. AUC means area under the receiver operating characteristic curve. Mean AUC was 0.93312 for a plain template. It was 0.93504 for an AR(16) background model with a 1 percent covariance floor. The paired gain was 0.00192 AUC. Its descriptive 95 percent person bootstrap interval was -0.01558 to 0.02083. The exact paired sign-flip p was 0.859375. This test did not show a clear gain. The background transform reduced later-control correlation, but this did not ensure better detection. Fixed secondary results depended on the floor and time window. The package includes exact voltage inputs, all scores and fitted weights, and three figures. It also includes a conditional score proof, code, source hashes, and a public retrieval recipe. A full replay matched all numerical results. Two complete public source files were freshly checked. The other source checks are inherited. This reused cohort does not establish a true neural response label or a live BCI benefit. This version fixes the file audit for a package downloaded with sci fetch. All numerical results are unchanged.
stereoelectroencephalography · SEEG · brain-computer interface · signal detection · autoregressive model · matched filter · AUC · electrical stimulation
References
6- supersedesAn AR(16) model with a 1 percent floor gave no clear 20–500 ms SEEG detection gain after five calibration pulses in eight epilepsy sessions sci:sha256:e8bb4cbe1d49abf394773608f1b7d8b27123ed920a24a659cbb86022cd602b50
This version fixes the file audit after sci fetch adds its publication metadata folder. The audit now omits that folder. All data, scores, models, figures, and numerical results are unchanged.
- usesPast pulse updates reduced normalized 20–500 ms SEEG forecast error versus fixed five-pulse templates at 86 selected sites in eight epilepsy sessions sci:sha256:fd07147e53abc31f50ad87cee1d777a9d91b45b3d2a172703485899005f00530
This publication supplied the exact all-pulse voltage windows, event joins, fixed readout choices, source metadata, retrieval code, and inherited source audit. We kept its early-selected readouts and tested a new background detector on every later good pulse.
- usesFive calibration pulses gave mean later signed template AUC of 0.939 compared with 0.859 for one pulse in eight SEEG sessions sci:sha256:1d2aa103275ad7400befec4c899740ede8f2eb59a09897bdc3f4d4e190e54418
This publication defined the original first-five energy rule and supplied the readout choices through the all-pulse parent package. We preserved those choices and read its selection code. No later detector score selected a readout.
- usesdoi:10.18112/openneuro.ds004696.v1.0.1
Version 1.0.1 supplies the processed SEEG derivatives used for the reported analysis.
- doi:10.1523/JNEUROSCI.2201-22.2023
- doi:10.1109/TASSP.1983.1164156
Cited by
6 publicationsPublic catalog entries on Sciscape.
- usesDonor correlations gave 0.005 lower mean AUC than local correlations with eight features and five fit pulses at 86 SEEG sites in eight patients
Its temporal whitening result motivated direct evaluation of response reading. This study tests donor feature correlations with a different reader. It does not make a numerical comparison with that model.
- usesA fixed spatial noise rule gave no clear 20 to 500 ms AUC gain after five early pulses in eight SEEG sessions
The earlier temporal noise test helped define the spatial noise question. No detector scores or fitted models from that study enter the current analysis.
- usesVariance weights from five baseline pulses gave no clear AUC gain for eight cosine features at 86 SEEG sites in eight patients
Used its full temporal noise-weighting study as context for the question and scope. The present rule instead estimates diagonal noise variances in eight fixed cosine features. The studies share the clinical cohort and are not independent confirmations.
- usesDonor covariance raised mean 16-mode SEEG response AUC after five local fit pulses at 86 selected sites in eight patients
The earlier local autoregressive background-reader result informed the choice to test shared temporal background covariance. The model and later-pulse set differ, so this is not a replication of that test.
- usesThresholds from other patients gave a 10.2% mean background call rate at a 5% target after five fit pulses at 86 selected SEEG sites in eight patients
Read its comparison of noise weighting and plain templates to choose a threshold calibration question. Its results were not included in the current outcome calculations.
- usesSix checks gave no clear AUC gain over one 100 ms check in eight SEEG sessions after eight calibration pulses
This publication supplied all 1042 good-pulse voltage windows at 86 fixed readouts. It also supplied metadata, hashes, retrieval code, and inherited audits. We used these inputs to test six prefix checks with separate control thresholds.