Random pair splits overestimated SEEG spectral-feature reliability relative to shaft splits

In 197 adult epilepsy patients in the RAM study, we compared split-half reliability of three patient-mean spectral features using 24,108 same-shaft bipolar SEEG pairs from archived 60 s word-list memory task spectra. After group quadratic MNI location adjustment, random pair splits gave Spearman rho values of 0.949–0.962, while splits by electrode shaft gave 0.814–0.871. Pair-split reliability exceeded shaft-split reliability by 0.091–0.135 (95% patient-bootstrap confidence intervals 0.069–0.119, 0.101–0.160, and 0.104–0.174; Holm-adjusted p=0.003 for each feature). Shaft-split full-implant Spearman–Brown reliability was 0.90–0.93. Without location adjustment, shaft-split rho was 0.81–0.84 and pair-split rho was 0.95–0.96. These findings show that pair splits can overestimate reliability when both halves contain pairs from the same shaft. The halves came from the same 60 s segment, session, amplifier, and site; the results therefore do not establish stability over time and also include state, hardware, and site effects. The 2–45 Hz slope was a simple OLS line in log-log space without peak removal, not a FOOOF exponent.

stereo-EEG · SEEG · reliability · split-half · spectral features · electrode shafts · RAM study