A prior fit without labels increased mean squared label error for a fixed HUP model in 18 external SEEG patients
We tested whether a prior fit without target labels reduced contact label probability error in public SEEG data. A fixed seven-input HUP model used 31 source patients from OpenNeuro ds004100 version 1.1.3. An intercept fit used source patient exclusion scores. The main test used 18 NIH and JHU patients from ds003876 version 1.0.2. A separate early prior fit increased mean patient Brier score from 0.09489 to 0.50924. The mean paired increase was 0.41435. Its 95 percent patient bootstrap interval was 0.24484 to 0.58336, conditional on the source fits and independent patient sample units. Error increased in 17 of 18 patients. Eight estimates reached the upper prior bound. AUC stayed the same. A later window with the early prior held fixed and an all-19 check gave the same direction of error change. This result concerns one fixed model, source probability scale, contact rule, and previously used cohort. It does not identify a failed model condition or measure surgical benefit. The package preserves code, compact spectra, labels, scores, patient errors, source versions and hashes, checks, and repeat instructions.
SEEG · epilepsy · probability calibration · prior probability shift · Brier score · external validation · OpenNeuro
References
9- 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
The parent preserves this source HUP training table, model rule, compact spectra, and Newton logistic solver. They supply source patient exclusion fits and an independent numerical fit check.
- usesAn intercept fitted to 15 development patients reduced mean SOZ label Brier score for a fixed balanced logistic model in 16 ds004100 SEEG test patients sci:sha256:0355c510cf9e19e9a2adfcdf302e677644eba54cddf24f46798319df38dd4bd5
Its fixed-model source intercept result informed the source fit rule. This study makes new source patient exclusion fits before the external comparison.
- usesReference choice reduced mean 80 to 150 Hz rank agreement more than time change in 19 public SEEG records sci:sha256:16d6a5e5126b007f6f6be4f042d829df32ccd751f41bcfda4c9b9a3f5a80fce2
The parent preserves this metadata selection rule, pinned receipts, EDF headers, parser, decoder, channel matching, and NumPy Welch code. They support a fresh three-record public spectrum check.
- usesAn odds correction reduced mean squared label error in 31 SEEG patient exclusion tests with mean features from two recordings. sci:sha256:41d62aa89527563522810258c871e83d6f158d3b015a05c9e177d18491073074
Its internal HUP probability correction result informed the source probability comparison. This study does not use its fitted predictions. It tests an external prior adjustment.
- usesWith independent patients and null sign symmetry, tests detected no HUP model SOZ AUC gain in 18 external SEEG patients sci:sha256:cacfcb73cf6624b3b7d324cad3700beeace003783bcff021dcba6e9769e783f7
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.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
Training data and fixed HUP model inputs came from this exact OpenNeuro dataset version.
- usesdoi:10.18112/openneuro.ds003876.v1.0.2
External patient data and labels came from this exact OpenNeuro dataset version.
- doi:10.1038/s41593-021-00901-w
The external dataset description requests this source paper for dataset acknowledgment; this study does not calculate neural fragility.
- doi:10.1101/2021.10.15.464594
The external dataset description requests this source-cited manuscript for dataset acknowledgment; this study does not calculate source-sink connectivity.
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.