With independent patients and null sign symmetry, tests detected no HUP model SOZ AUC gain in 18 external SEEG patients
We trained fixed logistic models on 31 HUP SEEG patients from public OpenNeuro ds004100. We then tested 18 NIH and JHU patients from ds003876. Test labels did not enter model fitting. The combined model used contact number and six CAR spectral features. Mean patient AUC was 0.7452, versus 0.7235 for contact number alone. The mean paired change was +0.0217. Its 95 percent patient bootstrap interval was [-0.0417, +0.0923]. The exact two-sided sign-flip p value was 0.553772. The interval conditions on the fixed training sample and models. The test requires independent patients and symmetric differences under the null. The interval includes zero. These data do not resolve a positive mean gain. The primary test used four 15-second windows from seconds 30 to 90. A later-window check gave change +0.0301. Source soz=n/a means unmarked, not normal. This result concerns public labels in sampled contacts. It does not establish a clinical decision rule. The package contains compact spectra, model coefficients, all contact scores, source pins, code, checks, and repeat instructions.
stereoelectroencephalography · seizure onset zone · external validation · area under the curve · spectral features · patient bootstrap
References
6- usesSix CAR spectral features raised mean seizure onset label AUC beyond contact number. Fixed logistic tests excluded each of 31 SEEG patients. sci:sha256:aee5a470b0833c4f856b9a317fa6e6005828d574f583119a67ef17a72b7a7b3f
We used the verified HUP contact table, six-feature model rule, original CAR spectra, and training retrieval code. They define the fixed training input and permit its repeat.
- usesReference choice reduced mean 80 to 150 Hz rank agreement more than time change in 19 public SEEG records sci:sha256:16d6a5e5126b007f6f6be4f042d829df32ccd751f41bcfda4c9b9a3f5a80fce2
We used the external metadata selection rule, pinned source receipts, EDF headers, decoder, and time checks. We retrieved new raw ranges and calculated new spectra for this transfer test.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
The fixed HUP training spectra and contact labels came from this dataset version.
- usesdoi:10.18112/openneuro.ds003876.v1.0.2
External patient recordings and source labels came from this dataset version.
- doi:10.1038/s41593-021-00901-w
- doi:10.1101/2021.10.15.464594
Cited by
5 publicationsPublic catalog entries on Sciscape.
- usesResidual 80 to 150 Hz concentration gave an uncertain mean AUC gain of 0.021 in 19 NIH and JHU SEEG patients after HUP fitting
Its related HUP-to-external spectral model and contact-number comparison informed this question. We used no outcomes from that model as observations in the new analysis.
- usesA prior fit without labels increased mean squared label error for a fixed HUP model in 18 external SEEG patients
The complete verified parent package supplies the fixed full HUP model, source and external spectra, contact rules, metadata, and pinned public retrieval recipes. This study tests probabilities in the same external patients.
- usesFor common-average spectra at 20 to 100 Hz, inverse-frequency weights changed top contact sets in 12 of 19 SEEG records
This work supplied the original 19 external spectrum files reused through the direct parent. We kept its contact pool and source intervals. We did not use its trained models.
- usesUnder patient independence and null sign symmetry, a test detected no mean 20–100 Hz slope change between averaging orders in 19 SEEG patients
This study uses its 19 saved test spectrum files, source code, and source receipts for the averaging comparison.
- usesPatient rank scaling did not show an AUC gain in 18 external SEEG patients under fixed model and site rules
We used its hash-verified compact spectra, fixed HUP and external samples, metadata, source receipts, and raw retrieval code. We tested a new patient rank model rule and resampled both training and test patients.