Understanding neurodegeneration in Down syndrome through AI-driven neuroimaging analysis

Down syndrome (DS) is characterized by distinctive neuroanatomical features and a markedly increased risk of early-onset Alzheimer’s disease (AD). However, the anatomical correlates of neurodegeneration in DS remain poorly understood, limiting early diagnosis and personalized interventions. This project aims to address this gap by developing AIbased tools to detect and quantify brain alterations in individuals with DS using 3D magnetic resonance imaging (MRI) data. We propose an unsupervised learning approach that does not rely on manual annotations. Specifically, we will use Variational Autoencoders (VAE) and Diffusion Models (DM) to identify morphological changes associated with AD progression in DS. These models will be trained separately on MRI data from euploid and DS individuals, enabling the creation of high-resolution 3D brain alteration maps. The latent representations learned by these models will then serve as input for lightweight machine learning classifiers to stratify DS individuals based on neurodegeneration severity. Building upon promising 2D pilot results, this study will extend analysis to 3D MRI, leveraging a large dataset of annotated scans. By combining advanced AI models with expert clinical evaluation, we aim to discover robust biomarkers and improve our understanding of how AD affects individuals with DS. Expected outcomes include accurate, non-invasive tools for early detection and monitoring of AD in DS, with broader applicability to other neurodevelopmental conditions. This interdisciplinary effort integrates expertise in neuroimaging, AI, and DS research, and is supported by strong preliminary data and institutional collaboration.