Across-session high-gamma channel maps are less similar than within-session maps in 8 SEEG subjects

We compared spatial maps of bipolar band power between two 30 s segments from the same recording session and segments from separate sessions in 8 subjects and 19 sessions from OpenNeuro ds005624. For high gamma (70–150 Hz), median across-session Spearman correlation across channels was 0.960, below the within-session median of 0.982 (7 of 8 subjects; exact one-sided sign-flip p = 0.0078). Across-session correlations nevertheless remained high (0.92–0.98). The other four bands showed the same direction, but their uncorrected p values (0.020–0.035) did not pass a Bonferroni threshold of 0.0125. Across-session channel identification from 1–150 Hz log-PSD profiles was above chance in all subjects (accuracy 0.21–0.47; chance 0.005–0.017; one-sided p = 0.0039), and lower than within-session accuracy (p = 0.0078). A single global offset explained a median 3% of high-gamma change and 3%–16% across bands. These results suggest that the coarse spatial pattern transfers, while global gain correction alone does not restore per-channel feature values. The study used eight subjects, fixed 30 s segments, and recordings with unknown dates and task state; its metric does not test task-evoked responses.

stereo-electroencephalography · SEEG · band power · high gamma · channel map · brain-computer interface · cross-session stability · spectral features