The hippocampus is a key structure involved in the early pathological progression of Alzheimer’s disease. Accurate segmentation of this region is vital for the quantitative assessment of brain atrophy and the support of diagnostic decision-making. To address limitations in current MRI-based hippocampus segmentation methods—such as indistinct boundaries, small target size, and limited feature representation—this study proposes an enhanced segmentation framework called FED-UNet++. The residual feature reconstruction block (FRBlock) is introduced to strengthen the network’s ability to capture boundary cues and fine-grained structural details in shallow layers. The efficient attention pyramid (EAP) module enhances the integration of multi-scale features and spatial contextual information. The dynamic frequency context network (DFCN) mitigates the decoder’s limitations in capturing long-range dependencies and global semantic structures. Experimental results on the benchmark dataset demonstrate that FED-UNet++ achieves superior performance across multiple evaluation metrics, with an IoU of 74.95% and a Dice coefficient of 84.43% ± 0.21%, outperforming the baseline model in both accuracy and robustness. These findings confirm that FED-UNet++ is highly effective in segmenting small and intricate brain structures like the hippocampus, providing a robust and practical tool for MRI-based analysis of neurodegenerative diseases.
Yang et al. (Tue,) studied this question.