BrainParc: Unified Lifespan Brain Parcellation with Anatomy-guided Progressive Transmission

Motivation: Accurate brain parcellation across lifespan is essential for understanding brain development, aging, and disorder. Existing parcellation algorithms are typically tailored to specific ages or data centers to address intensity variations from brain maturation and scanner difference, leading to inconsistent parcellation across lifespan. Goal(s): We propose a unified framework, BrainParc, for precise and consistent sMRI-based brain parcellation across lifespan. Approach: BrainParc first hierarchically extracts coherent, intensity-invariant anatomical (edge map) from sMRI, and then utilizes the edge map to progressively guide parcellation of brain tissues and regions. Results: Extensive experiments conducted on internal and external datasets consistently demonstrate superior performance and wide application of BrainParc. Impact: We present BrainParc, the first lifespan brain parcellation framework using a single model, and evaluate it on the largest known lifespan sMRI dataset to date (over 91.3 thousand scans), achieving highest precision and consistency than other segmentation models.