Multimodal Neuroimaging and Machine Learning Framework for Early Prediction of Autism Spectrum Disorder In Infants

Abstract

Early detection and subsequent diagnosis of autism spectrum disorder (ASD) is a critical measure to allow an individual to carry out effective interventions that enhance the development outcomes in the long term. The article under study examines the incorporation of multimodal neuroimaging and machine learning methods to forecast ASD in high-risk infants and in toddlers aged between 6 and 36 months. Features of cortical thickness, white matter integrity, and functional connectivity were extracted by using structural MRI, functional MRI and diffusion tensor imaging. Discriminative neural biomarkers were found with the use of advanced feature selection and dimensionality reduction algorithms and then classified with support vector machines, random forests, and deep neural networks. It was found that the proposed framework demonstrated a higher predictive performance than assessing behavior alone and identified early changes in long-range connectivity and localized structural differences as the predictors. These results illustrate the potential of integrating both the neurobiological indicators and the computational intelligence to support an earlier, objective, and individualistic approach to promoting ASD detection interventions