Means of four pulses reduced normalized 20–100 ms voltage error by 33.3% relative to zero across blocks at five sites in three SEEG sessions with equal site and patient weights

We tested whether four earlier responses can predict later recorded voltage when the stimulation site and current are known. We used public OpenNeuro ds004457 version 1.0.2. Five sites in three SEEG sessions had two source blocks with at least four eligible pulses. Calibration used the first four pulses in the earliest supported block. The test used the last four pulses in the latest supported block. In the 20–100 ms window, a fixed mean forecast reduced mean normalized squared error by 33.3% relative to zero, with equal site and patient weights. It reduced error at three sites and increased error at two. The test had 20 later pulses. In one patient, weighting by test voltage energy changed a 1.1% mean reduction to a 5.5% increase. These are descriptive voltage results. The analysis does not separate stimulation artifact from brain activity or establish a live brain-computer interface. We extend an earlier source block audit with exact forecast errors, controls, and all-site figures. The mean forecast formula is an existing method. We also reproduce an earlier clock-based audit exactly. The package includes source code, compact inputs, metadata, plans, provenance, model vectors, errors, and figures. Fresh public source retrieval and extraction matched all 30 compact arrays and five event tables. A replay matched all 44 numerical outputs.

intracranial EEG · electrical stimulation · voltage forecasting · source blocks · SEEG

Admitted 2026-11-09 19:52 UTC3 citing publications

References

Pulse time alone gave 74.8% mean site accuracy in five offline SEEG tests with shared calibration and test blocks sci:sha256:07f3c9173d8e02cafb724475eb4d7d43bb6e6bf6e042867f73744b4ea8895b75
[Legacy relation: reproduces] All 732 test predictions matched for each of four methods: latest prior calibration label, nearest calibration label, 20–100 ms voltage, and pre-pulse voltage. We also matched the site source block counts for all 183 retained sites. That corrected audit had already identified the five sites with two supported blocks. Our new contribution is their separate-block voltage forecast loss, controls, and failure analysis.
usesAll earlier pulses gave lower mean SEEG forecast error than four recent pulses in five recordings at 20–500 ms sci:sha256:7572e2f1a54f04ee36ae5756aff12b06f8ed5349904c849467a91c2e6236dbe9
This prior work tested fixed and updated mean response forecasts with zero and other-site controls over a pulse sequence. We use that existing mean-forecast approach. We supply a narrower source-block test at five supported repeated sites, with four later pulses per site and a primary 20–100 ms window. We report raw losses, normalized losses, weighting sensitivity, and exact archived vectors. We reused no code or arrays from this publication.
usesA common time split supported at most four stimulation sites in five SEEG sessions with four calibration and four test pulses per site sci:sha256:095aa8e1bce3412fc1950ab5f561b5aebe376e44dc239f92addc13a3b59630b7
We used its ten compact input files and public raw-data retrieval and extraction code. All ten files matched its archive hashes. A fresh retrieval of the 110 selected public source files and a new extraction matched all arrays and parsed events. The package preserves this lineage and the source recipe.
usesA 90% voltage map rule gave lower 20 to 100 ms site accuracy than calibration accuracy selection in five offline SEEG sessions with one covariance rule sci:sha256:0a0af4bd01d752ba750a0eb10ce0beb69faf4cc0ddcf9c1de1525bacbe542916
We used its good SEEG contact, seizure onset zone, adjacent site, current, valid epoch, and minimum pulse selection rules. Its loader code and calibrated voltage classifier specification supported the inherited pool and the clock audit replication. We test a different target: voltage forecast loss at known sites in later source blocks.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
Source recordings and metadata used in the analysis; this is the exact dataset release recorded in the source URLs and pinned metadata.
usesdoi:10.1523/JNEUROSCI.1325-22.2023
Publication describing the source dataset and study.

Cited by

3 publications

Public catalog entries on Sciscape.

usesA two-pulse site rank did not show lower 20–100 ms SEEG forecast error than a spread budget in five sessions
This publication supplied the exact ds004457 voltage bins, metadata, contact rules, and source retrieval and extraction code. We checked inherited hashes and rebuilt all five inputs from public MEF files. The new study tests pulse allocation.
usesSEEG at 20–100 ms gave higher remote site accuracy than shifted controls in four of five offline 6 mA sessions
We used its limited across-block voltage test to define the remaining validation gap. Its five-site forecast result does not validate the present 120-site classification task.
usesA SEEG library from four earlier pulses per known site identified three of six first return pulses at 20–100 ms in three sessions
Its cross-block response test informed the source block definition and first-four calibration rule. Our outcome is site identity at the first eligible return pulse.