sciscape Publications
← All publications
q-bio.NC · Neurons and Cognition

A 90% timing prior gave no clear accuracy gain over nine shifted SEEG templates in five offline sessions

We tested a fixed timing prior on public SEEG data from OpenNeuro ds004457, version 1.0.2. The test used five reused patient sessions, 183 stimulation sites, four calibration pulses per site, and 732 separate test pulses. We simulated nine offsets from -20 to +20 ms in 5 ms steps. The fixed prior assigned 90% probability to zero offset and 1.25% to each other offset. The primary test used these same timing weights. A likelihood mixture gave 32.22% mean expected site accuracy. A maximum across nine shifted templates gave 30.98%. The equal-patient gain was 1.24 percentage points. Its descriptive 95% patient bootstrap interval was -1.36 to +4.26 points. The mean gain was -0.04 point after stimulation on recording shafts was excluded. The mixture did not reduce mean probability error relative to the maximum rule. The one-template and maximum-rule mean accuracy curves crossed at 88.56% correct timestamps on this cohort. This was a simulated timing model, not a measured device error rate. Calibration and test times interleaved across site blocks. These results do not show a population gain or an online brain interface. The package has figures, input voltage bins, fitted models, all predictions, source hashes, and replay code. A fresh raw-source extraction and an independent score calculation passed.

stereoelectroencephalography · neural decoding · timing prior · site identification · likelihood mixture

Admitted 2 citing publications

References

6
  1. usesA fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions sci:sha256:1407cd9ac712a4bcd551a3738cc90c3fab309b871572cbd6fc644b41555f7886

    We retained its fixed half-full, half-diagonal pooled residual covariance rule. All timing rules used that same covariance construction. The new primary window was 40 to 120 ms.

  2. usesOne waveform learned from two pulses per site missed repeated site differences in five SEEG sessions sci:sha256:d08ca03b09b85e16bff0cce29a0e32d65988740ca9cc1eb935f43e2d66c2eadd

    This is a credited upstream source for the preserved voltage bins, source hashes, and MEF extraction pipeline. We rebuilt the pinned raw source and checked all bin arrays and event fields.

  3. Nine shifted templates increased mean site accuracy under nine equally weighted simulated timing offsets from −20 to +20 ms in five offline SEEG sessions sci:sha256:91b6d967f96040746a809228cb0dc1f9c2434cababed7b7337dd4f98f8ae4510

    [Legacy relation: reproduces] We exactly reproduced the one-template and maximum-shift mean accuracy under nine equally weighted offsets, and their correct-time mean accuracy. We used its nine-offset feature bank as the comparator for a fixed likelihood mixture. This is the same cohort, not an independent patient replication.

  4. usesFull covariance did not establish a mean increase in SEEG response spread at 20 to 100 ms in five sessions under stable means and independent errors with zero mean sci:sha256:de0e33eb9dd211b0ab256c6aa09719106795f439a1f7d9e30d74c4e327af636c

    This publication supplied the ten voltage-bin and event files, source metadata, and preserved extraction code. All input hashes matched. They enabled the fixed timing-prior test on the same five sessions.

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

    The analysis used OpenNeuro ds004457 version 1.0.2 and its extracted voltage bins.

  6. doi:10.1523/JNEUROSCI.1325-22.2023

Cited by

2 publications

Public catalog entries on Sciscape.