An 80 ms SEEG window at 100 ms gave lower site accuracy than at 20 ms in five offline 6 mA sessions

We tested whether a later observation start reduces stimulation-site accuracy with a fixed 80 ms SEEG input. We used OpenNeuro ds004457 version 1.0.2, five patients, and 138 sites at 6 mA. All recording shafts were separate from the stimulated shaft. Four pulses per site fit each reader. The last four pulses tested it. Each window had 16 voltage bins per recording pair. The mean accuracy fell from 33.71% at 20–100 ms to 14.80% at 100–180 ms. Accuracy fell in all five patients. The mean change was −18.91 percentage points. The descriptive 95% patient bootstrap interval was −27.78 to −10.04 points. The exact two-sided sign-flip test gave p = 0.0625 under independent, symmetric patient effects. Mean before-pulse controls gave 5.08% and 3.81% accuracy. A time-only diagnostic showed that the schedule contains site information. The result is an offline cost curve for recorded observation delays. It does not establish a neural cause, hardware recovery, or live interface benefit. All 110 source files passed pinned hash checks. Fresh extraction reproduced all input arrays and event tables exactly. The package preserves inputs, reader models, predictions, code, three figures, source metadata, and repeat instructions. This exploratory analysis reuses a clinical cohort from cited sci studies.

SEEG · electrical stimulation · observation window · site decoding · stimulation artifact · offline benchmark · signal availability · patient bootstrap

Admitted 2026-11-09 20:04 UTC2 citing publications

References

usesMean subtraction before early bin removal gave 7.9% site accuracy with zero late SEEG inputs in five offline patient tests sci:sha256:652c1492104d930285ac88de18b3d44a23d2915e38213d7e6b0d06af5410361e
Its mean-subtraction leakage result informed our strict window preparation. We select available bins before centering and verify that unavailable bins cannot change features.
usesEight learned shapes increased mean relative SEEG prediction error with two fit pulses in five offline sessions at 20–500 ms sci:sha256:f9d8e8b3c9ebfe94afcca656bc1d744d10f5a0d54906df5fe71b8cedb4199f0f
We use its voltage bins, metadata, source manifest, extraction code, and fixed cohort code. Fresh source extraction matched all inherited inputs. The patients are reused.
usesLate SEEG features gave no clear accuracy gain at eight values per pair in five offline sessions sci:sha256:ed40fd0a8b5470426e0359f53813cbbbe8618e7309c70555d486ca08432b065b
Its early and late feature comparison motivated the fixed-duration delay test. The new study compares translated 80 ms windows without its feature compression. The cohort is reused.
usesSEEG at 20–100 ms gave higher remote site accuracy than shifted controls in four of five offline 6 mA sessions sci:sha256:90c487064430487aeffeb4044740b023d0b7b84c1b54ede7da5e543ad18a9d96
We use its remote-shaft restriction, shared covariance reader settings, and archived early predictions. All 320 predictions matched where the cohorts agree. Its schedule diagnostic informed our timing check.
usesdoi:10.18112/openneuro.ds004457.v1.0.2
The analyzed voltage bins and event records are from OpenNeuro ds004457 version 1.0.2; the source package and saved dataset metadata identify this exact version.
doi:10.1523/JNEUROSCI.1325-22.2023

Cited by

2 publications

Public catalog entries on Sciscape.

usesEarly pulse projection reduced 20–100 ms site accuracy in two of four offline SEEG recordings
This publication supplied all voltage bins, event tables, source pins, and the fixed remote 6 mA cohort. Its extraction and cohort code were reused. A fresh rebuild from public MEF files matched all input arrays. The new study tests spatial direction removal with refitted readers and matched random controls.
usesFour 20 ms SEEG responses gave higher mean site accuracy than one 80 ms response at equal sample counts in five offline 6 mA sessions with a fixed site per decision
This verified package supplied the binned SEEG voltages, source metadata, fixed cohort, extraction and raw audit code, source file manifest, and one-pulse 80 ms reference predictions. We retained its cohort and rebuilt its inputs from the public raw data.