fMRINet: Repurposing the EEGNet model to identify emotional arousal states in fMRI data

In recent years, functional magnetic resonance imaging (fMRI) transformed our understanding of the intricate relationship between emotions and the brain. The precise classification of emotional states from fMRI data poses challenges for traditional machine learning methods dealing with high-dimensional data. The limitations of these conventional approaches have spurred a growing interest in exploring the potential of deep learning (DL) models. This study introduces a novel approach to classifying emotional arousal levels using fMRI data, specifically tailored for projects with limited data. The approach involves the adaptation of the EEGNet architecture, originally designed for the classification of electroencephalography (EEG) signals, to fMRI data. By mapping the fMRI signal into brain regions using a brain atlas, fMRINet is applied to the two-dimensional fMRI time series, achieving a promising performance in identifying emotional states in both typical and clinical participants (balanced accuracy between 70% and 72%). Our findings highlight the successful integration of the EEGNet architecture into fMRI data and contribute to the broader field of brain state classification.