The Investigation of Emotion Cognition in Autism Spectrum Disorder: A Multi-dimensional and Multi-stage Path Analysis

Emotion cognition deficits are a core component of Autism Spectrum Disorder (ASD) and remain difficult to capture due to their dynamic and cross-stage nature. This study proposes a multimodal stage-aware structural equation modeling (MS-ASEM) framework for interpretable emotion cognition assessment in ASD by integrating overt behavioral responses and covert physiological states. Specifically, this framework employs synchronized eye-tracking (ET), electroencephalography (EEG), electrodermal activity (EDA), and facial expression (FE) data to quantify four stages of emotional processing: perception, understanding, arousal, and expression. A structural equation model is developed to uncover cross-stage dependencies and atypical pathways in ASD. Experimental results with 40 children diagnosed with ASD and 50 typically developing children demonstrate a unique “high-arousal, low-integration” profile in ASD, characterized by hypersensitivity to negative stimuli, atypical prefrontal activation, and desynchronized multimodal responses. The proposed path model further reveals impaired transmission between perception and understanding, as well as inhibitory coupling between arousal and expression. Overall, the framework effectively bridges computational assessment with structural interpretability, highlighting the advantages of stageaware interpretable multimodal modeling in mechanism-oriented affective computing research.