The 95 percent interval with patient bootstrap refits included zero gain from six spectrum features at 88 SEEG label boundaries in 26 patients

We tested whether six fixed spectrum features add ranking information at consecutive SEEG contacts with different seizure-onset labels. We used public OpenNeuro ds004100 version 1.1.3. Each test patient was excluded from model training. The training cohort had 29 patients. The boundary test had 88 pairs in 26 patients. Mean patient concordance was 0.798 for contact number and 0.905 for contact number plus spectrum features. The observed gain was 0.107. Its 95 percent interval was -0.014 to 0.197 after 2,000 patient bootstrap draws with model refits. The interval included zero. The interval for all-contact AUC gain was 0.035 to 0.137 in the same patients. The point gain was positive in seven patients, negative in one, and zero in 18. Source coordinate units conflict with the coordinate values. Thus, the pairs define boundaries in contact names and clinical labels. They do not establish physical tissue boundaries. We verified 62 public EEG prefixes and reproduced all used spectrum arrays exactly. The package preserves data, source hashes, code, model scores, bootstrap results, and reproduction steps. This is an internal secondary analysis of one public cohort.

SEEG · epilepsy · spectrum features · patient bootstrap · boundary concordance · logistic regression · OpenNeuro