Clinical 2D MRI radiomics for infant brain age and trajectory modeling
TL;DR
A data-driven radiomics framework that integrates automatic registration, segmentation, and feature extraction for the analysis of routine T1-weighted and T2-weighted scans from children aged 0–2 years is established, transforming conventional 2D MRI data into a powerful, objective tool for early neurodevelopmental assessment.
Abstract
Summary Clinical 2D magnetic resonance imaging (MRI) is the cornerstone for assessing early brain development, yet the lack of a systematic methodology renders the fine-grained and quantitative evaluation of early brain growth underdeveloped. We address this gap using a data-driven radiomics framework that integrates automatic registration, segmentation, and feature extraction for the analysis of 2,893 routine T1-weighted and T2-weighted scans from children aged 0–2 years. The framework yields accurate brain age prediction (mean absolute error = 1.19 months) and pinpoints the “corpus callosum median” (CC-median) as the most critical, biologically relevant predictor. We used the generalized additive model for location, scale, and shape (GAMLSS) to model the normative trajectory of CC-median. This standardized curve robustly differentiates developmental delays from typical peers, achieving an accuracy of 0.96 (for ages 6–24 months). This work establishes a clinically compatible framework, transforming conventional 2D MRI data into a powerful, objective tool for early neurodevelopmental assessment.