A novel cross-layer dual encoding-shared decoding network framework with spatial self-attention mechanism for...

Area of Science:

• Neuroimaging
• Medical Image Analysis
• Artificial Intelligence in Medicine

Background:

• Accurate hippocampus segmentation in brain MRIs is vital for understanding neurodegenerative diseases like Alzheimer's.
• Current methods struggle with the hippocampus's small size, complex shape, and low contrast, often neglecting spatial information and feature correlations.

Purpose of the Study:

• To develop a novel deep learning framework, ESDSA, for enhanced hippocampus segmentation in human brain MRIs.
• To improve segmentation accuracy by incorporating spatial self-attention and a cross-layer dual encoding-shared decoding network.

Main Methods:

• Proposed a novel cross-layer dual Encoding-Shared Decoding network with a Spatial self-Attention mechanism (ESDSA).
• Integrated spatial self-attention to capture crucial spatial information of the hippocampus.
• Designed a dual encoding-shared decoding network to extract global MRI and hippocampus spatial features.

Main Results:

• The ESDSA framework achieved a Dice Similarity Coefficient (DSC) of 89.37% on T1-weighted structural MRI data.
• The Spatial Self-Attention (SSA) and dual Encoding-Shared Decoding (ESD) strategies individually improved DSC by 9.47% and 5.35% compared to baseline U-net.
• ESDSA demonstrated superior performance over state-of-the-art methods for hippocampus segmentation.

Conclusions:

• The proposed ESDSA framework effectively enhances hippocampus segmentation accuracy in brain MRIs.
• The integration of spatial self-attention and cross-layer dual encoding-shared decoding strategies are key to ESDSA's superior performance.
• ESDSA offers a promising approach for neuroimaging research, particularly in the study of neurodegenerative disorders.