Brain Age Estimation: A Multi-Region Approach Using Groupwise Registration and 3D Shape Contexts Derived From Displacement Vector Fields

Brain age estimation has emerged as a crucial biomarker for quantifying inter-individual variability in brain aging and identifying risks of cognitive decline. This is particularly relevant for the study of neurodegenerative conditions like Alzheimer’s disease, where deviations from normal brain aging patterns can serve as early indicators of pathology. We introduce an interpretable framework for brain age estimation that adopts a multi-region perspective to capture heterogeneous neuroanatomical changes. Groupwise registration was employed to construct sex- and age-stratified templates, enabling the computation of anatomically meaningful displacement vector fields (DVFs) that characterize voxel-wise structural deviations from the population average. From these DVFs, features were extracted in fifteen brain regions using a 3D shape context descriptor to encode displacement direction and magnitude. Validation was performed on 1,956 T1-weighted MRI scans from cognitively normal individuals in the ADNI dataset (ages 60–90). The combined multi-region model achieved a mean absolute error (MAE) of $1.66 \pm 0.15$ years ( $R^{2} = 0.84 \pm 0.05$ ) for males and $1.81 \pm 0.12$ years ( $R^{2} = 0.80 \pm 0.04$ ) for females, significantly outperforming many contemporary deep learning models. These findings demonstrate that biologically interpretable, registration-derived features—when analyzed from a multi-regional perspective—can yield robust and accurate estimates of brain age. Beyond predictive accuracy, the framework provides insights into the regional specificity of aging processes, thereby offering a foundation for early detection strategies and targeted clinical applications in neurodegenerative disease research.