A fixed covariance rule gave higher SEEG site accuracy than a diagonal rule in offline tests from five patient sessions
We tested whether correlation between SEEG features helps identify a known pulse site. We used OpenNeuro ds004457 version 1.0.2 and all five patient sessions. We kept 183 sites and 15 recording pairs. Four early pulses at each site trained fixed templates and residual covariance. Four later pulses at each site tested the rules. Mean accuracy was 40.6% with half covariance and 34.6% with diagonal variance alone. The mean gain was 6.0 percentage points, with positive gains in all five patients. The exact two-sided patient test gave p = 0.0625. The descriptive patient bootstrap interval was 3.1 to 8.8 points. This small sample does not establish a population effect. Calibration and test times interleaved across sites. The result is an offline comparison within sessions, not an online calibration test. Fixed window and contact checks retained positive gains. A later exploratory check suggests that covariance across time bins accounts for much of the gain. We rebuilt the inputs from 110 public source files and verified all hashes. A repeat run matched all 825 saved model arrays. The package preserves three figures, pulse predictions, fitted models, source code, data provenance, and reproduction instructions.
SEEG · stimulation site decoding · covariance · regularized discriminant analysis · brain-computer interfaces · intracranial electrophysiology · offline benchmark
References
4- usesA joint SEEG shape decoder did not show a clear mean gain over one selected pair in four patient sessions sci:sha256:81f77d4f52365d5b14fed7d3d6db7b430a2cf478b9a6a6062540c4e90bbd8e7c
We used this report’s verified voltage bins, source metadata, retrieval and extraction code, pulse eligibility, chronological per-site allocation, and equal-pair shape comparator. We rebuilt its inputs from raw files. Our covariance test uses the same patients and sessions, so it is not independent replication.
- usesSEEG response shapes at 15–100 ms gave higher site identification accuracy than shapes at 100–500 ms within each of five patient sessions sci:sha256:aa5b71d76d214f861e99964287f9da01f3f9fc6c707a4dcfe24807283918ab40
We used this earlier site benchmark to choose the pulse-site identification question and response-window controls. It uses the same source patients. We did not treat its evidence as an independent sample.
- usesdoi:10.18112/openneuro.ds004457.v1.0.2
The analysis used OpenNeuro ds004457 version 1.0.2; the source data and metadata record CC0.
- supportsdoi:10.1523/JNEUROSCI.1325-22.2023
The source study supplies the published context that temporal and limbic stimulation produces distinct responses in ventral temporal cortex.
Cited by
112 publicationsPublic catalog entries on Sciscape.
- usesCovariance-scaled rank-four maps gave 35.1 percent site accuracy versus 32.6 percent for voltage maps at 20 to 100 ms in five SEEG sessions with four calibration pulses per site
This source supplies the original compact-bin extraction method and fixed covariance method. The present study rebuilds its bins from the pinned MEF source and checks every array.
- usesA SEEG library from four earlier pulses per known site identified three of six first return pulses at 20–100 ms in three sessions
We reused the fixed five-session pool, 183 sites, 15 recording pairs, 5 ms contact bins, raw extraction code, and fixed covariance settings. We checked all 1,904 eligible pulses and rebuilt all input values from pinned raw files. The new test uses a growing library of completed earlier responses.
- usesLinear maps fitted to pulse responses had more 20 to 100 ms SEEG forecast error than background maps in five selected sessions at ridge strength 0.1
We used its selected contacts, voltage-bin inputs, original metadata, raw-source manifest and extraction code to build the matched forward-time forecast benchmark. We checked the input hashes and rebuilt all bins from the pinned public source.
- usesA fixed SEEG distance rule accepted most pulses from unlearned stimulation shafts in five sessions
We used its fixed shared covariance rule and known-site decoder context. The new code tested full response distance for rejection of pulses from unlearned stimulation shafts.
- usesTwo Mahalanobis distance limits raised mean known-site rejection more than mean unknown-shaft rejection for 20–100 ms voltage in five offline SEEG sessions at matched nominal 80% acceptance
The study uses the fixed covariance rule with a 0.5 diagonal weight. It does not select a new weight from test results.
- usesRemoving missing features raised mean site accuracy for 20 to 100 ms responses under known simulated contact loss in five SEEG sessions
We used its source-contact bin arrays, event tables, recording eligibility code, and fixed 0.5 covariance rule. These supplied the intact model for the new known contact-loss test. All intact predictions matched this parent.
- usesNoise correction did not increase mean eight-site SEEG accuracy at 20–100 ms with four calibration pulses per site in five sessions
We used its exact voltage-bin arrays, event tables, contact and candidate rules, and raw extraction and retrieval code. We rebuilt the inputs from the pinned public raw source and obtained exact agreement. The present analysis adds a matched ordinary-versus-noise-corrected site-selection test. The cohort is shared and does not give independent evidence.
- usesCovariance terms changed mean offline SEEG balanced accuracy by −0.09 points with 16 calibration trials in eight recordings
Its matched covariance and diagonal comparison for stimulation sites motivated this question. This run uses task and rest power from another dataset and an automatic Ledoit-Wolf covariance estimate. It does not replicate the stimulation-site result.
- usesShared SEEG templates used 36% fewer numeric values with a 0.22% mean relative error decrease in five offline 20 to 100 ms tests
We used its extraction and contact inclusion code through the direct input package. The same raw recipe rebuilt the compact inputs in this run.
- usesEight learned shapes increased mean relative SEEG prediction error with two fit pulses in five offline sessions at 20–500 ms
This is the original archive source of the binned voltage inputs and fixed recording contacts used through the response-shape publication.
- usesTwo Mahalanobis distance limits raised mean known-site rejection more than mean unknown-shaft rejection for 20–100 ms voltage in five offline SEEG sessions at matched nominal 80% acceptance
The study uses the fixed covariance rule with a 0.5 diagonal weight. It does not select a new weight from test results.
- usesRank limits gave no site labels at a 90 percent coverage target across five reading times in five offline 6 mA SEEG sessions
Its compact voltage bins, fixed recording contacts, MIT retrieval code, and extraction code reach this study through the direct parent. The source files and hashes are preserved. This run rebuilt the inputs from public raw recordings.