Adding high gamma raises mean AUC by 0.020 in patient holdout tests of a contact order model within named shafts in 33 HUP SEEG patients
We tested whether high gamma power adds to contact order when ranking clinical seizure onset contacts within named shafts. We used two 60 second interictal records per patient from OpenNeuro ds004100 version 1.1.3. We fitted the models with each test patient held out. In 33 patients, mean AUC was 0.7816 for contact order and 0.8018 when power was added. The mean gain was 0.0202. Its 95 percent patient bootstrap interval was -0.0200 to 0.0537, with full model refitting. The interval includes zero. Twelve patients improved, thirteen had lower AUC, and eight tied. The gain was 0.0079 after omission of HUP180. The evidence does not establish a population gain or equal model performance. We preserve source covariance data, patient predictions, all bootstrap draws, checks, and raw retrieval code. This is a secondary analysis of shared HUP patients. It tests agreement with supplied clinical labels.
stereoelectroencephalography · high gamma power · contact order · seizure onset · patient holdout · AUC · epilepsy
References
3- usesLower contact numbers gave mean seizure onset AUC 0.794 within mixed inferred shafts in 34 selected ds004100 SEEG patients sci:sha256:50d777142ef4ebc45c41203ba318cb2244c6b807e8d402b44d8ba5067784feec
This result supplied the contact order benchmark idea and led to the new incremental model question. Its numerical result files were not model inputs. The studies share source patients.
- usesMean high gamma AUC within inferred shafts exceeds a label control in one-minute SEEG samples with an average reference from 33 HUP patients sci:sha256:46a12570ac1f98c21c7f72a6958c443edd3986f81154a8c30cc11c1f760e42a9
This work supplied the exact covariance inputs, source code, and pinned raw metadata. We verified all copied file hashes. We used these inputs to test the added value of power after contact order with patient holdout fitting.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
Source dataset: HUP iEEG SEEG records and clinical contact labels from the exact analyzed OpenNeuro ds004100 v1.1.3 snapshot.
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesThe interval for added STE HFO rank gain included zero in fixed contact order models for 17 SEEG patients
This work supplied the design for fixed pairwise ridge models, contact order control, patient holdout tests, and patient resampling with model refits. The present study applies the design to SEEG HFO counts in a different cohort. Its code and numerical results were not imported.
- usesTime concentration gave no clear AUC gain for 80 to 150 Hz SEEG power with fixed linear models and independent resampling of 16 test patients
We use this earlier contact number and power benchmark to state the scope of the new contribution. We read it after our new model results. The new test adds time concentration to an existing power and contact number base model. It does not introduce the contact number benchmark.
- usesThe patient bootstrap interval for AUC gain from low frequency SEEG power above contact order included zero in 19 external patients after HUP training
Its comparison of power with contact order motivated the test of added signal information. We test a different frequency band and external patients.