2.5D HAU-Net with gated spatial attention for automatic hippocampus segmentation in MRI.

Area of Science:

• Medical Imaging
• Neuroscience
• Artificial Intelligence

Background:

• The hippocampus is crucial for Alzheimer's disease (AD) diagnosis.
• Accurate automated segmentation of the hippocampus is vital for analysis.
• Current computational limitations hinder clinical application of CAD systems.

Purpose of the Study:

• To develop an efficient and accurate automated hippocampal segmentation framework.
• To enhance feature selectivity and robustness in segmentation models.
• To provide a resource-efficient solution for clinical deployment.

Main Methods:

• A 2.5D U-Net-based framework (HAU-Net) integrating an attention mechanism.
• Input representation by stacking three consecutive MRI slices for enhanced spatial context.
• Utilizing a gated spatial attention module and a hybrid Dice-BCE loss function.

Main Results:

• Achieved high Dice scores of 91.05% (MSD Task04) and 90.62% (HarP).
• Demonstrated superior performance and generalization compared to baseline U-Net and other models.
• Maintained computational efficiency suitable for practical clinical use.

Conclusions:

• The attention-guided 2.5D HAU-Net offers an effective and robust solution for hippocampal segmentation.
• Its low computational demand and strong performance support clinical integration.
• The framework is promising for neuroscience, medical imaging, and AD research.