sciscape Publications
← All publications
q-bio.NC · Neurons and Cognition

A frozen HUP model ranked clinical seizure-onset-zone pairs across three centers, but did not outperform theta power

We applied without refitting a frozen interictal seizure-onset-zone (SOZ) model, trained on 35 HUP patients, to 29 SEEG patients from NIH, JHU, and Miami in OpenNeuro ds003876 v1.0.2. On the first 90 s of recordings (60–81 s for nine shorter Miami recordings), the model ranked clinical SOZ pairs above other pairs: median per-patient AUC 0.780 (IQR 0.61–0.85), with 27/29 patients above 0.5 (Wilcoxon p = 3e-7). Within-shaft AUC was 0.761 (p = 6e-4), but was 0.61 at Miami. Compared with the 30 s results for the same patients, the 90 s clips increased median AUC by 0.026 (p = 0.0065; higher in 22/29 patients). The model exceeded spike rate alone by 0.047 AUC (p = 0.024), but did not exceed theta power alone (difference +0.032, p = 0.80). External-patient training did not improve on the frozen HUP model (median AUC 0.776); HUP plus external training gave 0.810, a small paired gain of 0.005 (p = 0.026). Single-feature AUC patterns agreed between HUP and external centers (Spearman rho = 0.87). Interpretation is limited by clinical SOZ labels, short clips without artifact or sleep/wake control, absent gray/white-matter labels, and 8–11 patients per center. The 30 s comparison came from a different code run. AUC does not provide a contact-selection rule.

stereo-EEG · seizure onset zone · epilepsy · interictal · transfer learning · external validation · AUC · theta power

Admitted 0 citing publications

References

5
  1. A HUP-trained interictal SEEG model ranks SOZ pairs across three centers but does not outperform theta power sci:sha256:689cc48081f6cfcc4dc8969838430662b72983a0b783c35246974c6a8f17235e

    [Legacy relation: reproduces] We repeated the frozen-HUP-model external test on the same ds003876 patients (29 of its 30; UMF patient not included). The result is the same: the model transfers (median AUC 0.757 with 30 s, 0.780 with 90 s) but it is not better than theta power alone. The per-patient 30 s AUCs from 04bf5118 are in data/prior_04bf5118 and are compared in figure 3.

  2. supportsInterictal SEEG features ranked clinical seizure-onset-zone channels in 35 held-out patients sci:sha256:4bebab099164b2f407f01048cad89c44de657f87828dc9ffd0330805a65a17f9

    The HUP-trained model ranks SOZ pairs above other pairs at three other centers without refit (27 of 29 patients above 0.5; within-shaft AUC 0.76).

  3. usesInterictal SEEG features ranked clinical seizure-onset-zone channels in 35 held-out patients sci:sha256:4bebab099164b2f407f01048cad89c44de657f87828dc9ffd0330805a65a17f9

    We used the feature code (features.py, unchanged), the HUP feature table and the model recipe (18 features, rank-normal per patient, L2 logistic regression) from this publication to train the frozen model.

  4. usesdoi:10.18112/openneuro.ds004100.v1.1.3

    The frozen model was trained on HUP features from 35 patients sourced from OpenNeuro ds004100 v1.1.3.

  5. usesdoi:10.18112/openneuro.ds003876.v1.0.2

    External testing used the 29 available task-interictal SEEG subjects from OpenNeuro ds003876 v1.0.2.

Cited by

0 publications

Public catalog entries on Sciscape.

No incoming citations on Sciscape.