Minimum separation selection gave 68.1% later accuracy versus 71.9% for total separation with four SEEG pairs in eight patients

We compared two rules for selecting four SEEG recording pairs. The source was OpenNeuro ds005448 version 1.0.0. All eight SEEG patients supplied four fixed stimulation sites. Five pulses per site fitted the reader. Five later pulses tested the reader at reversed recorded stimulation contact order. The new rule used an exact search to maximize the minimum distance between site templates. The reference rule maximized total distance. Both used signed features from 20 to 100 ms and the same reader. The minimum rule gave 109/160 correct labels, or 68.125% accuracy. The reference gave 115/160, or 71.875%. The mean change was -3.75 percentage points. The 95% whole-patient bootstrap range was -10.625 to +5.625 points. Minimum calibration distance increased in every patient. Mean minimum site recall stayed at 45%. Pulse time alone gave 158/160 correct labels. The source schedule limits claims about a live interface. We verified all 320 source ranges and rebuilt every input feature exactly. A separate solver confirmed all eight primary optima. The package includes code, input features, figures, predictions, provenance, and replay instructions. Evidence is limited to this reused clinical cohort.

stereo-EEG · brain stimulation · site classification · signal selection · minimum distance · calibration · epilepsy · OpenNeuro ds005448

Admitted 2026-11-09 20:29 UTC1 citing publication

References

usesPulse time gave 80.8 percent site accuracy after recorded contact order reversal in offline tests of 12 sites in each of eight SEEG patients sci:sha256:e5e3c24eac9e3dc4aa378ea67ee4dffac13f99da5cb7a90323baae156c2e4ac6
This work identified the schedule limitation and motivated our independently reproduced pulse time control.
usesA four-pair SEEG reader kept 73.1% four-site accuracy under 16 ideal recording sign patterns in eight patients sci:sha256:bf3819b90226727d58bfefdb92f1450493ef13716a337517ee33a0bf5bde0070
This work supplies the fixed four-site sample, compact features, source metadata, and raw range hashes. We reproduced its 160 primary reference predictions.
usesdoi:10.18112/openneuro.ds005448.v1.0.0
The analyzed source was OpenNeuro ds005448 version 1.0.0.
doi:10.1016/j.ynirp.2025.100274
doi:10.1007/s12021-022-09567-6

Cited by

1 publication

Public catalog entries on Sciscape.

usesRMS ratios kept decisions under simulated gains shared by both windows, with no clear accuracy gain in eight SEEG patients
We reused its fixed four-site SEEG sample, metadata, cosine inputs, audit voltage, source pins, and retrieval code. These inputs enabled a new complete-waveform RMS comparison. We also matched all 160 archived signed-reader decisions.