UniSurf: Universal lifespan cortical surface reconstruction

Accurate and efficient reconstruction of cortical surfaces from MRI throughout the lifespan is essential for mapping normal and abnormal brain development, maturation, and aging, and for facilitating early diagnosis of neurodevelopmental and neurodegenrative disorders. Existing methods, however, often lack generalizability and longitudinal consistency in lifespan data, owing to pronounced variations in intensity and contrast arising from brain maturation, aging, and differences across MRI scanners, hindering their utility in both research and clinical settings. To address this, we present UniSurf, a universal framework for lifespan cortical surface reconstruction. Notably, UniSurf leverages intensity- and contrast-invariant anatomical information along with a differentiable iso-surface extraction algorithm, enabling robust, volume-surface-consistent brain tissue segmentation and cortical surface reconstruction across the lifespan. Moreover, we curate a largest MRI dataset to date, comprising 10,330 T1-weighted MRI data from 14 globally distributed cohorts spanning ages 0-100 years, to train and evaluate UniSurf. Extensive experiments on this large-scale, heterogeneous dataset demonstrate that UniSurf significantly outperforms state-of-the-art methods in lifespan cortical surface reconstruction, generalizing across diverse populations and imaging conditions and achieving average Chamfer distances of 0.317 mm for white-matter surfaces and 0.467 mm for pial surfaces. Beyond accurate cortical surface reconstruction, UniSurf enables reliable, spatially accurate, longitudinally consistent tracking of cerebrum volume, cortical surface area, and mean cortical thickness across the lifespan, highlighting its potential as a robust and generalizable tool for large-scale cross-sectional and longitudinal studies and clinical translation.