A fixed donor model raised 20–500 ms SEEG error for simulated shaft loss after five fit pulses in eight people

We tested recovery of a hidden SEEG shaft after electrical stimulation. We used public ds004696 data from eight people, 82 sites, and 874 recording-pair instances. Five early pulses fit each model. The final five pulses tested it. A fixed model used a scalar from other shafts in the same 20–500 ms window. Mean normalized squared error rose from 0.8136 for a scaled template to 1.0624 for the donor model, a 30.6 percent increase. Error rose in all eight people. The person bootstrap interval for the relative increase was 20.4 to 42.5 percent. This is an offline result for simulated known shaft loss. It does not test future prediction or a live brain-computer interface.

SEEG · electrical stimulation · evoked potentials · waveform recovery · shaft loss · brain-computer interface

Admitted 2026-11-09 20:29 UTC3 citing publications

References

usesShared removal gave mixed 20–500 ms SEEG forecast effects and raised error on shafts with simulated gain 4 faults after five calibration pulses in eight people sci:sha256:ad4550515ed194f9bc18e3881bd88d07b0dfe8c748ea6d00542ff843c07eecdf
We used its exact pulse products, BIDS metadata, pinned source hashes, and extraction code. We added a same-pulse scalar donor model for recovery after known simulated shaft loss. We downloaded and checked the complete public sources for two people in this run.
supportsA fixed joint SEEG model increased mean normalized error compared with separate scaling after five calibration pulses at 20–500 ms in eight sessions sci:sha256:073d5f5841c5def5608c980eec11bdcff16306c43558f2fb5c0fdf46cc72d8f8
Its fixed joint template had more mean normalized error than separate scaling after five fit pulses in the same cohort. Our new test also favors the scaled template, but tests a different input: signals from other shafts on the same later pulse. The two tests are not independent cohorts.
usesA before-pulse energy model gave 1.1 percent more SEEG error forecast loss after five fit pulses at 85 selected sites in eight people sci:sha256:bc8499bf519fb55735568b16bcd97dfc3abc90e3f7c70687a2a3d0df66ce206b
We read this related before-pulse forecast result to define the scope of the new test. It used one selected recording pair per site. Our test uses a panel of remote shafts to recover a hidden same-pulse waveform. Its fitted models and waveform subset are not inputs here.
usesdoi:10.18112/openneuro.ds004696.v1.0.1
The analysis uses compact pulse products derived from OpenNeuro ds004696 version 1.0.1.
doi:10.1523/JNEUROSCI.2201-22.2023
doi:10.1111/epi.17528

Cited by

3 publications

Public catalog entries on Sciscape.

usesCalibration range limits reduced 20 to 500 ms donor SEEG error at penalty one after five fit pulses in eight people, but a fixed template had lower error
This supplied the fixed scalar donor feature, relative ridge slope, and gain projection. Its range-exceedance result motivated the new clipping comparison. Our raw-derived cohort and all-later split differ from its processed-data cohort.
usesCalibration range limits reduced 20 to 500 ms donor SEEG error at penalty one after five fit pulses in eight people, but a fixed template had lower error
This supplied the fixed scalar donor feature, relative ridge slope, and gain projection. Its range-exceedance result motivated the new clipping comparison. Our raw-derived cohort and all-later split differ from its processed-data cohort.
usesA fixed contact geometry term raised mean SEEG prediction error by 0.21% at 20 to 500 ms in eight tests that each excluded one person
We use the exact ten-pulse Gram products, pulse maps, source metadata, source hashes, and GPL extraction code. They permit all eight held-person size-ratio tests without new recordings.