Pre-stimulation high-gamma/theta power did not predict recall or modify the 50 Hz stimulation effect in 11 FR2 patients

We tested whether the recall cost of open-loop 50 Hz stimulation during word encoding depended on brain state before stimulation began, as proposed for closed-loop memory stimulation. In 27 sessions from 11 patients in OpenNeuro ds005489, we measured log power from 1.0 to 0.25 seconds before each word pair. A training-free high-gamma/theta tilt (80–140 Hz minus 3–8 Hz) did not predict recall in control pairs (AUC 0.495). The stimulation effect was similar in low- and high-tilt pairs: the high-minus-low difference was −1.2 percentage points (95% CI −12 to +12; sign-flip p = 0.87). A logistic classifier produced a +16 percentage point contrast (p = 0.004), but its cross-validated control-pair scores were anti-correlated with recall (AUC 0.40), making this contrast an artifact rather than valid evidence of state dependence. The results are limited by low power, 11 of 38 patients, and only 3 patients with mostly SEEG contacts.

memory · free recall · intracranial EEG · closed-loop stimulation · brain state · high gamma · theta power · state dependence · classifier bias