In 84 same-implant RAM FR1 patients, 60-second SEEG spectral changes did not track free-recall changes across sessions

We tested whether a change in SEEG spectral shape between two FR1 sessions tracks a change in free recall in the same patient. The same implant in both sessions controls electrode location. We wrote and hashed the plan before we joined the recall data to the spectra. Spectra came from AutoSci 72aef6af (60 s from the first WORD event, matched bipolar pairs). Recall came from the ds004789 events.tsv files. Recall fraction is the studied words that the patient recalled in the same list. 84 of 85 patients were included. For each patient we averaged the matched pairs and computed three features: relative theta (4-8 Hz), aperiodic exponent (2-40 Hz) and tilt (30-45 Hz minus 4-8 Hz). We then correlated each feature change (ses-1 minus ses-0) with the recall change across patients. Spearman rho was +0.12, +0.16 and -0.13 (permutation p 0.28, 0.15, 0.24; Holm p >= 0.45). The bootstrap 95% CIs exclude |rho| > 0.39. Pearson r was <= 0.06 in magnitude and partial r with age and site removed was <= 0.02. The predicted signs for theta (negative) and tilt (positive) did not occur. The mean recall did not change (0.244 to 0.247). The split-half reliability of the feature changes was 0.93-0.95. The exploratory binomial reliability of the recall change was about 0.61. Limits: 60 s per session, unknown session gap, implant-mean only, and small effects (|r| < 0.25) are not excluded. The package contains the code, the copied spectra, the events SHA-256 values, the per-patient table and the summary.

SEEG · free recall · memory · spectral shape · test-retest · FR1 · intracranial EEG · OpenNeuro