GA-Shape: A Dual-Path Framework Guided by Gestational Age and Local Shape Descriptors for Fetal Brain MRI Segmentation

Accurate segmentation of fetal brain tissues in magnetic resonance imaging (MRI) is essential for diagnosing diseases, assessing developmental delays, and analyzing patterns of fetal brain development. The rapid growth of the fetal brain during pregnancy results in significant shape variations at different gestational ages, while the folding of the cerebral cortex introduces subtle local shape changes, making brain tissue segmentation more challenging. To tackle these challenges, we present a dual-path framework for brain tissue segmentation, which integrates a primary segmentation network with an auxiliary network designed to predict gestational age and local shape descriptors (LSDs). The predicted gestational age and LSDs are integrated into the outputs of the encoder block in the segmentation network through a fusion block, transforming the spatial features and enhancing the ability of the network to capture complex anatomical structures and tissue boundaries. We evaluated the framework on the FeTA2021 dataset, achieving a Dice similarity coefficient of 88.36%, demonstrating its effectiveness in fetal brain tissue segmentation.