Search Papers | Poster Sessions | All Posters
Poster B156 in Poster Session B - Thursday, August 8, 2024, 1:30 – 3:30 pm, Johnson Ice Rink
Disentangling human amygdala activity with artificial neural networks
Grace Jang1, Philip A. Kragel1; 1Emory University
Human neuroscience has revealed that neuroimaging measures and neural activity in the amygdala encodes a wide array of variables ranging from threat, salience, valence, and stimulus intensity. Although much has been learned about amygdala function, an overarching account of amygdala processing remains elusive. Here we use a combination of human neuroimaging, computational models of visual processing, and self-report measures of emotional experience to develop and validate encoding models that predict patterns of amygdala response acquired with fMRI during movie-viewing. When tested on naturalistic emotional images, we found that amygdala encoding models predicted ratings along the dimension of valence. Moreover, we found that the encoding models could be paired with deep generator networks to synthesize artificial stimuli that specifically engage the amygdala and anatomically defined amygdala subregions. These findings establish an approach for advancing our understanding of amygdala function by identifying how the amygdala transforms rich sensory inputs into low-dimensional representations relevant for behavior.
Keywords: amygdala encoding models artificial neural networks fMRI