Semi-supervised Fetal Brain Parcellation via Hierarchical Learning Framework

Abstract

Automatic parcellation of fetal brain regions using magnetic resonance (MR) images has become a valuable tool for studying prenatal brain growth and development. However, manual segmentation on large-scale fetal brain images is challenging, leading to limited annotated data availability. Although previous works have made progress, they are limited by not considering hierarchical nature and complementary information between brain regions. To overcome this limitation, we introduce a novel method to hierarchically segment the fetal brain into 87 distinct regions. The method employs a three-level coarse-to-fine network with the coarse level providing prior information to aid the fine level for fine segmentation. The first level predicts 8 brain regions, the second level refines the first-level 8 brain regions into 36 regions, and the final level refines further into 87 regions. This design hierarchically decomposes the fine difficult-to-achieve segmentation task into the coarse relatively-easy-to-achieve tasks by using guiding information from coarse level. Additionally, we introduce a data augmentation module to simulate variations in imaging conditions. To ensure robust segmentation performance under diverse imaging conditions, the network is trained in a semi-supervised manner using simulated data combined with a small set of labeled real data. In this way, we address the issue of limited high-quality labeled data, and enhance the model’s robustness to MR scanner variability. Extensive experiments on 558 neonatal subjects from the dHCP dataset and 176 fetal brain MR images demonstrate excellent segmentation performance of our method in terms of Dice score (91.42%), outperforming the second best nnUNet (88.77%).