Autism Res. 2025 Aug 23. doi: 10.1002/aur.70105. Online ahead of print.
ABSTRACT
Autism spectrum disorder (ASD) is characterized by impaired social interaction and communication skills, with semantic processing difficulties being a hallmark feature that significantly impacts social communication. While traditional neuroimaging studies have provided insights into language processing in ASD, ecological validity remains a challenge, particularly when assessing young children. This study introduces a novel approach to evaluate atypical semantic processing in children with ASD (aged 4-10 years) through electroencephalography (EEG) data collection during cartoon viewing, offering a more natural assessment environment. We developed an innovative methodology combining pretrained language models with regression techniques in a machine learning framework. The analysis incorporated the Six-dimensional Semantic Database system and EEG topographical mapping to investigate semantic processing preferences and neural mechanisms across various word dimensions. Our semantic processing model demonstrated robust performance with high sensitivity (91.3%) and moderate specificity (61.0%); findings successfully replicated in validation analysis. These results reveal distinct patterns in how children with ASD process semantic information, particularly in their integration and response to emotional semantic dimensions. These findings help us understand the language processing patterns in ASD and provide potential applications for auxiliary diagnosis in more natural settings, meeting important needs in clinical practice.
PMID:40847596 | DOI:10.1002/aur.70105
