Researchers Use AI and Brain Imaging to Decode Continuous Emotional States from Neural Activity

Researchers developed a novel framework that treats emotion decoding as a multi-target regression problem tracking continuous emotional trajectories over time, departing from traditional discrete classification approaches. They leveraged large language models to automatically extract fine-grained sentiment profiles from an auditory narrative and used these as proxies for subjective emotional states measured via fMRI in human subjects. The study employed regularized and kernel-based machine learning algorithms to analyze dynamic functional connectivity—how different brain regions interact over time—rather than static activity in individual regions. Models trained on temporal snapshots of dynamic connectivity significantly outperformed traditional region-of-interest approaches, effectively capturing how emotions fluctuate with narrative input. Using explainable AI techniques, the researchers identified emotion-specific patterns of brain network organization, providing interpretable insights into which neural connections drive emotional responses. The results support psychological constructionist theories suggesting emotions emerge from distributed network interactions rather than localized brain regions.