An added intercept reduced mean SOZ label probability error in 31 SEEG patient exclusion tests with a fixed logistic model
We tested label probabilities from a fixed class balanced logistic model in public SEEG data. The data were OpenNeuro ds004100 version 1.1.3: 31 patients and 2,879 eligible contacts. SEEG means stereoelectroencephalography. SOZ means seizure onset zone. Each test excluded its patient from model fitting and probability adjustment. The model used contact number and six spectral features from 30-second signal samples. An added intercept reduced equal-patient mean Brier error from 0.2047 to 0.0886. The reduction was 0.1161. Its approximate 95 percent patient bootstrap interval was [0.09125, 0.13845] with 300 complete nested refits. Mean predicted fraction fell from 0.4197 to 0.1045; the observed mean fraction was 0.1045. Each patient's ranking was unchanged. A model without class balancing gave similar error. The refit interval for the advantage over a constant label fraction included zero. This is an internal audit of source SOZ labels. It does not establish tissue causation or clinical benefit. The package preserves code, spectra, metadata, retrieval hashes, all predictions, fits, resamples and checks.
stereoelectroencephalography · seizure onset zone · probability error · Brier score · logistic regression · patient exclusion · class balancing · nested intercept
References
5- 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 its fixed seven-input class balanced logistic design. Our new endpoint was probability error in excluded patients, with a nested added intercept.
- usesAdding six spectral features increased mean seizure onset label recall within a 10 percent contact budget in 31 SEEG patients sci:sha256:0f6cce692e40a83c6b13de0829228520cb54bdf0669c03e23fd9c77d721dafe1
This package supplied the fixed 31-patient cohort, recording features, solver and complete source archive. We rebuilt the features and added a probability audit.
- usesIn 30-second HUP SEEG samples, reference changes reduced mean 30–100 Hz slope rank agreement more than run changes sci:sha256:b9c7b4646dba303292f492835f91dbb8026d4f35fcce938519d1d3fa6bcbff63
Its public SEEG spectra, EDF code, quality audits, metadata and pinned retrieval records are preserved in the direct source archive and support this analysis.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
The analysis used the HUP iEEG Epilepsy Dataset, OpenNeuro ds004100 version 1.1.3, for the 31-patient SEEG cohort.
- doi:10.1093/jamia%2Focac093
Cited by
1 publicationPublic catalog entries on Sciscape.
- supportsAn odds correction reduced mean squared label error in 31 SEEG patient exclusion tests with mean features from two recordings.
The same score error pattern occurred with features averaged across both recordings in the same 31 patients. Its separate-recording scores are preserved and independently recalculated for a descriptive comparison. This is a scope extension, not an independent cohort replication.