Machine-learning based prediction of future outcome using multimodal MRI during early childhood.

Abstract

The human brain undergoes rapid changes from the fetal stage to two years postnatally, during which proper structural and functional maturation lays the foundation for later cognitive and behavioral development. Multimodal magnetic resonance imaging (MRI) techniques, especially structural MRI (sMRI), diffusion MRI (dMRI), functional MRI (fMRI), and perfusion MRI (pMRI), provide unprecedented opportunities to non-invasively quantify these early brain changes at whole brain and regional levels. Each modality offers unique insights into the complex processes of both typical neurodevelopment and the pathological mechanisms underlying psychiatric and neurological disorders. Compared to a single modality, multimodal MRI enhances discriminative power and provides more comprehensive insights for understanding and improving neurodevelopmental and mental health outcomes, particularly in high-risk populations. Machine learning- and deep learning-based methods have demonstrated significant potential for predicting future outcomes using multimodal brain MRI acquired during early childhood. Here, we review the unique characteristics of various MRI techniques for imaging early brain development and describe the common approaches to analyze these modalities. We then discuss machine learning approaches in predicting future neurodevelopmental and clinical outcomes using multimodal MRI information during early childhood, highlighting the potential of identifying biomarkers for early detection and personalized interventions in atypical development.