A fixed spectral random forest had higher mean AUC with contact tests than patient tests in 36 SEEG patients
We tested a fixed spectral forest on public SEEG data from 36 patients. The analysis used 4,252 good contacts and two 120-second records per patient. Mean patient AUC was 0.748 with contact tests and 0.665 with patient tests. The paired difference was 0.084. Its descriptive 95 percent interval was 0.054 to 0.116. Each contact's two records stayed together. No contact appeared in training and test data. Shaft tests, separate records, common average inputs, and logistic regression provided planned checks. The grouping gap varied by model and patient. The package preserves source data, provenance, code, all assignments, predictions, and reproduction checks. The intervals condition on the fitted models. This single-cohort result does not establish clinical benefit.
SEEG · seizure onset zone · spectral features · AUC · patient-separated testing · contact testing · random forest · model validation
References
4- usesAverage references reduced low frequency seizure onset ranking AUC in two short SEEG recordings from each of 36 patients sci:sha256:ac5e50a0fe3911b4b562b31f3ee7fb2ab9caf8fe9695b3ad882f2c66ce6e9f33
The archived spectral measurements and fixed contact cohort supplied inputs for new contact, shaft, and patient test comparisons.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
OpenNeuro ds004100 version 1.1.3 supplied the SEEG recordings and clinical labels analyzed here.
- doi:10.1038/s41598-019-42347-y
Background study motivating the slow and delta spectral bands; its band, recording conditions, and protocol differ from this study.
- doi:10.3389/fnins.2024.1373515
Background study on dependence between EEG training and test segments; this study asks a distinct-contact SEEG grouping question.
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesA fixed logistic model had lower mean seizure onset ranking AUC with two added power variation features in 18 SEEG test patients
This publication supplied the four spectral feature conditions and the matched mask for 4,252 contacts. We rebuilt the features from source powers and tested the added value of two variation features.
- usesThe 95% interval for SOZ AUC gain from patient ranks included zero for a fixed logistic model in 18 test SEEG patients
We used its fixed contact cohort and four spectral model inputs. We verified all 409 archived files and preserved its complete package. We added patient rank scaling and a fixed test patient group.
- usesA fixed spectral model predicted random patient labels above chance across separate contacts and named shafts in 36 known SEEG patients
Used the four-feature contact input recipe and independent raw check code. Added random patient labels, equal patient weights, complete saved predictions, and fixed contact, shaft, and patient tests.
- usesAdding power variation lowered mean seizure onset ranking AUC in a fixed linear model of two short SEEG records from each of 36 patients
Its spectral model study informed the fixed median power and variation features, patient weighting, and whole-patient test split. The new analysis isolates the added value of variation in a fixed linear model.