A 60 s SEEG spectrum linked 21 of 65 patients across days; electrode positions linked 62 of 62

We tested whether a short stereo-electroencephalography (SEEG) recording could identify the same patient in recordings from another day. A pre-registered analysis compared session-1 probes from 65 patients with session-0 spectra from 261 RAM patients. For each recording, the signature was the median log10 power across accepted bipolar pairs at 87 frequency bins from 1.5 to 44.5 Hz in one 60 s segment; channel names were not matched. Pearson correlation ranked candidates. The primary analysis identified 21 of 65 patients (accuracy 0.32, 95% Wilson CI 0.22–0.44; chance 1/261 = 0.0038; one-sided exact p = 9.2e-35; median rank 3; top-5 rate 0.55). Among same-site candidates, 39 of 65 were identified (0.60; mean chance 0.032; exact p = 1.7e-44, Holm-adjusted p = 3.4e-44). After adjustment for an MNI location prediction, 25 of 62 probes were identified among 255 candidates (0.40; chance 0.0039; exact and Holm-adjusted p = 8.8e-44). The 1–8 Hz and 8–30 Hz bands each identified patients; the 30–45 Hz band did not (p = 0.22). An exploratory same-site, same-sampling-rate analysis identified 42 of 65 (0.65; mean chance 0.053; p = 1.2e-43). MNI pair positions alone matched 62 of 62 probes (median distance to the true match 0.05 mm), because coordinates were the same in both sessions. These results show a stable patient-specific spectral component and linkage risk from both signal and electrode coordinates. The signal result persisted after location adjustment. The study covers one implant, one montage, one hardware set, one task (FR1), and RAM patients; it did not test linkage across implants, tasks, or centers. The number of days between sessions is not reported in the source works. There were 65 probes, and a trained method could perform better; reported signal accuracies are a lower bound on linkage risk. For these data, removing labels alone does not prevent session linkage.

stereo-electroencephalography · SEEG · patient re-identification · cross-session linkage · spectral signature · privacy · electrode coordinates