Window power normalization gave an onset AUC change interval that included zero for 30–80 Hz bipolar SEEG in 31 patients
We tested whether window power normalization changed seizure onset ranking in public SEEG data from OpenNeuro ds004100 version 1.1.3. The primary test used two 60 s interictal excerpts per patient, 553 adjacent bipolar pairs, 71 shafts, and 31 patients. The fit and normalization used 30 to 80 Hz, with 55 to 65 Hz removed. Positive pairs had released onset marks on both contacts. Negative pairs had no onset mark on either contact. Each shaft had both classes. We gave patients equal weight. Mean onset AUC was 0.6561 for the ordinary mean spectrum and 0.6450 for the normalized mean spectrum. The paired change was -0.0111, with a patient bootstrap 95 percent interval from -0.0392 to 0.0195. The interval included zero and changes in both directions. This result does not establish equal performance. On the same pairs, mean patient median absolute slope differences between files fell from 0.2720 to 0.1919, a 29.4 percent reduction. The paired change interval was -0.1455 to -0.0317. Better file agreement did not establish better onset ranking in this test. All eight fixed secondary conditions are reported. The package preserves the plan, source metadata, pinned signal retrieval recipe, spectra, code, results, and reproduction steps. All 72 source excerpts were retrieved again. The full raw rebuild reproduced all 504 arrays and all result files. A separate reader and spectral check passed on 24 pairs at four sample rates. The result is conditional on the selected label and shaft rules. We did not test surgical outcomes or clinical benefit.
stereoelectroencephalography · SEEG · spectral slope · window power normalization · seizure onset · AUC · file agreement
References
4- usesWindow power normalization reduced mean file slope differences in paired 60 s bipolar SEEG records from 36 patients, with a 30 to 80 Hz fit and 55 to 65 Hz removed sci:sha256:b4ffa9fee5cbcff9ecff350d036ae6ff20b55d4c2948fe860337f5adb5b1ea90
We used its frozen 36-patient SEEG cohort, normalized and ordinary spectra, source metadata, pinned signal retrieval recipe, and raw processing code. These files supported the new paired onset ranking test and the complete raw rebuild.
- usesBipolar reference gave higher mean onset AUC than CAR for 30 to 80 Hz SEEG pair slopes within shafts in 31 patients with stable, matching contact marks sci:sha256:2db828e8f20176a7a8f07fdc95c3c0ddefa2018f85db4df6edaa6568b376de70
We used its stable endpoint label rule and within-shaft AUC rule. Its contact labels and ordinary bipolar AUC results supplied independent reference checks. The current test adds the window normalization comparison.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
The analysis uses the HUP iEEG dataset, version 1.1.3, for its SEEG cohort and recorded source data.
- doi:10.1038/s41593-020-00744-x
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesWith independent, exchangeable patients, added 30–80 Hz slope gave a loss gain interval that included zero after contact order adjustment in 31 SEEG patients
We used its fixed cohort, spectra, clinical marks, metadata, source pins, and signal code. The current study adds a contact order baseline, patient exclusion during training, and a refitted patient bootstrap. We retrieved all 72 source excerpts again and checked the full raw rebuild.
- usesAdding relative gamma power change did not raise within-shaft onset-label AUC in 15 test SEEG patients with two 60-second files
We used its exact paired window powers, retained contact lists, source metadata, retrieval recipe, and MIT signal code. Fresh raw retrieval reproduced all analysis inputs. These inputs enabled the new test of added information from gamma change between files.