In five bilateral patients, left-amygdala band power decoded aversive blocks better on average than right-amygdala band power

We tested whether bilateral amygdala recordings improve decoding of aversive versus neutral video blocks, using archived per-amygdala band-power features from OpenNeuro ds003374 (version 1.1.1) in five patients with recordings from both sides. Within-patient shrinkage LDA with leave-one-pair-out cross-validation gave mean block AUCs of 0.775 for the left side, 0.494 for the right side, and 0.725 with both sides concatenated. In four of five bilateral patients, right-side AUC was 0.33–0.46; the left side was better in those four. The prespecified endpoint, both-side AUC minus the mean single-side AUC, was +0.090 (4/5 positive; two-sided p = 0.125). The two-side model was lower than the left side alone by 0.05 on average. Cross-side transfer had mean AUC 0.653 (one-sided p versus 0.5 = 0.0625). Across all 14 amygdalae, mean single-side AUC was 0.85 for eight left amygdalae and 0.56 for six right amygdalae; this comparison is exploratory. With five bilateral patients, the smallest possible two-sided p-value is 0.0625, and no test reached p < 0.05. Side is confounded with electrode location, seizure-onset zone, and signal quality. The fixed-order block design does not separate emotion from video content. These estimates do not establish a general hemispheric effect or support bilateral implantation for this decoder; a larger sample is needed.

amygdala · SEEG · affective BCI · band power · LDA · laterality · cross-validation · video decoding