Mean and median rank disagreement marked greater loss of top-eight high gamma rank across two interictal SEEG records in 36 patients
We tested a calibration flag in public SEEG data from OpenNeuro ds004100 version 1.1.3. The sample had two 60 s interictal records for each of 36 patients. We selected eight bipolar pairs by mean power from 70 to below 110 Hz in record A. We flagged a selected pair when median power did not select it. In record B, 22 of 30 flagged pairs lost their top-eight mean-power rank. The count was 54 of 258 pairs without a flag. Adding the flag reduced patient-held-out Brier loss from 0.1953 to 0.1724. The mean gain was 0.0229. A patient bootstrap with model refits gave a descriptive 95% interval from 0.0041 to 0.0449. This added check followed the primary result. The flag caught 29% of rank losses. The package includes a calibration tool, power data, source records, code, and figures. These results test power rank stability in a reused epilepsy cohort. They do not establish task decoding or stimulation effects.
stereoelectroencephalography · SEEG · high gamma power · feature selection · rank stability · brain-computer interface
References
6- supersedesMean and median rank disagreement marked greater loss of top-eight high gamma rank across two interictal SEEG records in 36 patients sci:sha256:f8ad2bc0dd04577793e7f3f012ac31bcf2b45a5a4ec8f5c01e93b73b2f5d4877
Corrected the package file checker to ignore the metadata folder added by sci downloads. Data, scientific code, numerical results, and figures are unchanged. The revised checker passed with the full package and the added sci metadata folder.
- usesEight high-gamma SEEG pairs selected with ten seconds had 76.7% mean retention across two interictal records in 36 patients sci:sha256:4947f6220434c945bfb5f749fa56c65ce926ce8b7742925c1b030438e32ad73f
The two-record cohort and the earlier power-transfer question informed this test. The present analysis includes seizure onset pairs and tests a flag set from calibration power alone.
- usesMean and median power selected 61 and 55 SEEG pairs with seizure onset labels at eight pairs per patient in 60 s records A from 36 patients sci:sha256:700b53bb82755c0c922fc57119efeae9d7342793443bfb985bf6cd22309be61e
Used all 72 sixty-second power files, the matched 36-patient cohort, contact labels, native audit signals, source hashes, and retrieval code. The new test uses calibration rank disagreement to predict pair-level rank loss. The source set-overlap results were repeated as an audit. They are not independent evidence.
- usesdoi:10.18112/openneuro.ds004100.v1.1.3
Source data: HUP iEEG Epilepsy Dataset, OpenNeuro ds004100 version 1.1.3.
- doi:10.21105/joss.01896
- doi:10.1038/s41597-019-0105-7
Cited by
0 publicationsPublic catalog entries on Sciscape.
No incoming citations on Sciscape.