sciscape Publications
← All publications
stat.ML · Machine Learning

An odds correction reduced mean squared label error in 31 SEEG patient exclusion tests with mean features from two recordings.

The test used public SEEG data from 31 patients and 2,879 contacts. The features were averaged across two short recordings. Each test patient was absent from training. A fixed training odds correction reduced mean Brier error from 0.203 to 0.089. The mean reduction was 0.115. The patient bootstrap used 500 draws and refit both models. Its approximate 95 percent interval was [0.0939, 0.1341]. The correction kept mean AUC at 0.767. The corrected error was close to the error from natural patient weights. The gain over a constant fraction was 0.0054. Its interval was [-0.0010, 0.0093]. This result extends a same-cohort study that tested separate recordings. The result shows lower squared error for source seizure onset labels in this cohort. It does not establish calibrated probabilities or surgical benefit.

SEEG · stereoelectroencephalography · Brier score · odds correction · probability scores · patient-level validation · epilepsy

Admitted 1 citing publication

References

7
  1. 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

    This study uses its fixed 31 patient cohort, contact features, model, and saved scores. A new solver also reproduced all 2,879 primary balanced scores within 0.0000005.

  2. usesContact splits gave higher mean seizure-onset label AUC than patient splits for a fixed spectral random forest in 31 public SEEG patients sci:sha256:b6a361054e4ebea1dd4b8f14fdd99003ce2639da49c389eca05a7a1c82bba254

    Its preserved package supplies the source spectra, metadata, cohort audit, and raw signal retrieval recipes.

  3. supportsAn added intercept reduced mean SOZ label probability error in 31 SEEG patient exclusion tests with a fixed logistic model sci:sha256:a32fdd1ccc073ddf3db9bc65ce80932c0fdac591a3d9a9625aba86167f290b26

    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.

  4. 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 signal spectra are the underlying data for the preserved feature rows and independent signal checks.

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

    This study uses source SEEG signals and clinical contact labels from the exact OpenNeuro ds004100 version 1.1.3.

  6. doi:10.1016/j.jbi.2024.104666
  7. doi:10.1093/jamia%2Focac093

Cited by

1 publication

Public catalog entries on Sciscape.