Deep learning models enable high-performance, real-time emotion decoding from intracranial brain activity

Abstract

By integrating intracranial neural signals from gray and white matter, we developed personalized deep learning models that accurately decoded continuous valence and arousal states across image and video viewing tasks. The models generalized across tasks, revealed emotion-encoding brain subnetworks and enabled low-latency real-time decoding, bringing affective brain–computer interfaces closer to being suitable for closed-loop therapy.