Unveiling complex brain dynamics during movie viewing via deep recurrent autoencoder model

• Published in: NeuroImage (Orlando, Fla.) Vol. 310; p. 121177
• Main Authors: Wang, Kexin, Song, Limei, Li, Zhaowei, Wang, Liting, He, Xiaowei, Ren, Yudan, Lv, Jinglei
• Format: Journal Article
• Language: English
• Published: United States Elsevier Inc 15.04.2025; Elsevier Limited; Elsevier
• Subjects: Adaptability; Adult; Arousal; Autoencoder; Brain - diagnostic imaging; Brain - physiology; Brain Mapping - methods; Brain research; Cognition & reasoning; Deep learning; Dynamic large-scale brain networks; Emotions; Emotions - physiology; Energy consumption; Female; Functional magnetic resonance imaging; Humans; Information processing; Magnetic resonance imaging; Magnetic Resonance Imaging - methods; Male; Memory; Methods; Motion Pictures; Movie narrative structure; Naturalistic fMRI; Nerve Net - diagnostic imaging; Nerve Net - physiology; Neural networks; Neuroimaging; Recurrent autoencoder model; Sliding window strategy; Visual Perception - physiology; Young Adult; Naturalistic fMRI; Recurrent autoencoder model; Movie narrative structure; Sliding window strategy; Dynamic large-scale brain networks
• DOI: 10.1016/j.neuroimage.2025.121177
• ISSN: 1053-8119

•Leveraging naturalistic fMRI, we characterize the dynamic, complex interactions among large-scale brain networks during the viewing of emotionally charged movie.•The dynamic functional interactions covary with the development of the film's narrative, especially among the interactions of default mode network and dorsal attention network.•The dynamic interactions are also temporally synchronized with specific features of the movie, especially with the emotional arousal and valence, such as the interactions network associated with executive control network. Naturalistic stimuli have become an effective tool to uncover the dynamic functional brain networks triggered by cognitive and emotional real-life experiences through multimodal and dynamic stimuli. However, current research predominantly focused on exploring dynamic functional connectivity generated via chosen templates under resting-state paradigm, with relatively limited investigation into the dynamic functional interactions among large-scale brain networks. Moreover, these studies might overlook the longer time-scale adaptability and information transmission that occur over extended periods during naturalistic stimuli. In this study, we introduced an unsupervised deep recurrent autoencoder (DRAE) model combined with a sliding window approach, effectively capturing the brain's long-term temporal dependencies, as measured in functional magnetic resonance imaging (fMRI), when subjects viewing a long-duration and emotional film. The experimental results revealed that naturalistic stimuli can induce dynamic large-scale brain networks, of which functional interactions covary with the development of the film's narrative. Furthermore, the dynamic interactions among brain networks were temporally synchronized with specific features of the movie, especially with the emotional arousal and valence. Our study provided novel insight to the underlying neural mechanisms of dynamic functional interactions among brain regions in an ecologically valid sensory experience.