Knowledge-empowered Lifespan Neuroimage Analysis - The increasing availability of large-scale lifespan brain MRI data, from Fetal Tissue Annotation (FeTA), Developing Human Connectome Project (dHCP), National Database for Autism Research (NDAR), Baby Connectome Project (BCP), Multi-visit Advanced Pediatric (MAP) Brain Imaging Study, Autism Brain Imaging Data Exchange (ABIDE), Human Connectome Project (HCP), Adolescent Brain Cognitive Development (ABCD), Open Access Series of Imaging Studies (OASIS), Chinese Color Nest Project (CCNP), UK Biobank (UKB), Alzheimer’s Disease Neuroimaging Initiative (ADNI), and Australian Imaging Biomarkers and Lifestyle Study of Ageing (AIBL), affords unprecedented opportunities for precise charting of lifespan brain changes from the fetus to the elderly, providing important insights into the origins and aberrant evolving patterns of neurodevelopmental, psychiatric, and neurodegenerative disorders. However, a major barrier is the critical lack of lifespan-dedicated computational tools/pipelines for accurate and consistent processing and analysis of challenging lifespan brain MRIs, which suffer from age-dependent imaging artifacts and low and dynamic tissue contrast—especially for infant and elderly brain MRIs—as well as large inter- site data heterogeneity inherent to different imaging protocols and scanners across sites. Most existing tools/pipelines are designed for age-specific groups, and directly applying these to lifespan data results in inconsistent and misleading outcomes. Therefore, the overarching goal of this project is to create and disseminate the first knowledge-empowered deep learning pipeline for accurate and consistent atlas construction, skull stripping, artifact correction, super resolution, harmonization, tissue segmentation, topological correction, and brain labeling of lifespan brain MRIs. Specifically, we will develop an anatomy- guided atlas learning neural network to construct the first set of lifespan atlases with temporally consistent and spatially detailed patterns, and subject-specific atlas with personalized prior for subsequent aims (Aim 1). Guided by prior knowledge from Aim 1, we will develop a novel lifespan skull stripping framework (Aim 2). After skull stripping, we will then propose a novel deep learning framework for joint artifact correction, super resolution, harmonization, and tissue segmentation (Aim 3). We will propose an anatomically constrained topological correction network to ensure the topological correctness of cortex and further a transformer brain labeling to divide the cortex into anatomical/functional regions of interest (Aim 4). Finally, we will integrate our lifespan- dedicated tools and atlases into a comprehensive software package that is conveniently executable in both local and cloud-based environments (Aim 5). We will freely release our software, atlases, and processed lifespan data (150,000+ scans from 50,000+ subjects) to the public.
