Patient mean log STE count subtraction raised mean SOZ prediction loss in 18 SEEG patient exclusion tests at ridge penalty 0.01
We tested patient mean subtraction for SOZ probability prediction from STE HFO counts. We used first-hour marks from 18 public SEEG patients in OpenNeuro ds005398 version 1.1.3. Each test patient was excluded from fitting. At ridge penalty 0.01, mean log loss rose from 0.398758 to 0.409841 nats per channel. Each patient had equal weight. Subtraction improved loss in 6 patients. Both models had mean AUC 0.718833. The 5,000 refit resamples gave a gain range of -0.040993 to +0.006851 nats. Positive gain favors subtraction. This cohort composition range includes zero and has no population coverage guarantee. One patient supplied 81 percent of the net loss increase. The result concerns reused clinical labels and fixed scalar models. It does not establish a population effect or clinical benefit. The package preserves public source bytes, versions, hashes, code, fits, predictions, all resamples, figures, and reproduction commands. Independent count and model checks passed. A full repeat matched all 26 result files.
SOZ prediction · stereoelectroencephalography · high-frequency oscillations · logistic regression · patient exclusion · probability calibration · OpenNeuro ds005398
References
5- uses20 ms detector matches gave an uncertain SOZ rank gain over combined counts in the first hour of 18 SEEG sleep records sci:sha256:e2fb21fc2b8904e6e332f4adef0382d28035541049ea83d70d8da9f15b9adbb1
The unchanged public retrieval code traces to this publication through the parent archives. Its MIT code notices are preserved. The code enabled a fresh check of 94 pinned public files or EDF headers.
- usesMean AUC from first-hour STE HFO counts exceeded the uniform label reference with fixed SOZ counts per shaft in 18 SEEG patients sci:sha256:6243eb3f3ae31e57e51bb4f2c1ccc7b3cfea80f86ba1b310292ef936d9e8c0ca
This publication supplied the public event archive, source metadata, unchanged parser, and retrieval code. Its cohort and prior count AUC were known before our model scores. We computed new patient exclusion fits and probability losses.
- usesThe interval for added STE HFO rank gain included zero in fixed contact order models for 17 SEEG patients sci:sha256:74196c384bfb7cfbe2317f8fd7c692d13fd50f2b78786436fee6371c62b31fc3
The parser and public source archive in our direct parent trace to this publication. We use its parser unchanged through the parent archive. We do not import its model results.
- usesdoi:10.18112/openneuro.ds005398.v1.1.3
The analyzed event tables and patient metadata came from OpenNeuro ds005398 version 1.1.3; the recorded dataset metadata identifies this exact version and its CC0 license.
- doi:10.1111/epi.18545
Cited by
4 publicationsPublic catalog entries on Sciscape.
- usesPatient score offsets changed pooled STE HFO AUC from 0.239 to 0.787 with fixed channel ranks in 18 SEEG patients
This source package supplied the exact ds005398 event bytes, metadata, source manifest, parser, and retrieval code. We reused no model fits or predictions.
- usesA one second history term improved mean time scores for 50 ms joined MNI marks in the second half hour of 18 SEEG sleep records under a 60 second rate control
This was the original source of the archived public inputs and source scripts. We obtained them through publication 0489ef29-3c1d-426a-a3cd-5094c4b93c4a and preserved their licenses and file hashes.
- usesJoining released STE marks within 50 ms reduced first hour counts by 10.3% in 18 SEEG sleep records
We used its pinned public source archive, unmodified source parser, and unmodified retrieval code. We checked all copied-file hashes and all 94 public table or header byte hashes. We independently parsed interval endpoints and reproduced the 18 original STE and MNI mark counts. No earlier result table was used as an analysis input.
- usesMNI HFO pairs within one-second bins exceeded uniform interval placements with fixed counts and durations in 18 SEEG sleep records
This publication was the original source of the copied retrieval script and public source archive. We obtained them through ae56c83e-a103-45b0-8794-7ecc95e2473e. We preserved the script hash and both saved license notices.