Mean SEEG spectra identified 54% of 84 patients across two sessions

We tested whether a patient’s mean spectrum from matched bipolar SEEG pairs could identify that patient in a second session. Archived spectra from 84 RAM FR1 patients were analyzed over 2–40 Hz; both sessions used the same implant. Pearson correlation identified the correct patient in 53.6% of cases for the full spectrum and the offset-removed spectrum, and 54.2% after removal of an aperiodic line (chance 1/84, or 1.2%; permutation and Holm-adjusted p = 0.0003 for all three features). For the aperiodic-line-removed feature, mean self-correlation was 0.835 versus 0.534 for other patients; 80 of 84 patients had a positive differential-identifiability score. Within-session halves gave 71.4–78.6% accuracy. Per-frequency ICC(3,1) had a median of 0.69. Within-patient pair-level identification had median accuracy 0.20 and median chance 0.010, with 81 of 84 patients above chance. Because the same implant was used across sessions and electrode locations differed across patients, these results show patient-plus-implant specificity and do not establish a brain trait independent of electrode location. The planned offset-removed feature was redundant because Pearson correlation is invariant to subtracting a constant. One patient was excluded after a frequency-grid check that was not in the plan.

SEEG · spectral fingerprinting · patient identification · intracranial EEG · test-retest reliability · RAM FR1