MEA-Net: multilayer edge attention network for medical image segmentation.

Area of Science:

• Medical Image Analysis
• Deep Learning
• Computer Vision

Background:

• Medical image segmentation is crucial for diagnosis.
• Deep learning models, particularly encoder-decoder networks, are widely used.
• Existing methods often overlook low-level edge features, leading to imprecise segmentation boundaries.

Purpose of the Study:

• To propose a novel Multilayer Edge Attention Network (MEA-Net).
• To enhance medical image segmentation by effectively utilizing edge information.
• To improve the accuracy and reliability of diagnostic tools.

Main Methods:

• Developed MEA-Net, incorporating feature encoder, decoder, and edge modules.
• Integrated an edge feature extraction module to process multi-stage edge information.
• Employed a multilayer attention guidance module to refine edge features.

Main Results:

• Achieved high accuracy (up to 0.9993) and Dice coefficients (up to 0.9902) across four diverse medical image datasets.
• Demonstrated superior performance compared to state-of-the-art methods on standard evaluation metrics.
• Validated MEA-Net's effectiveness on tongue, retinal, lung, and clinical images.

Conclusions:

• MEA-Net significantly improves medical image segmentation by leveraging edge information.
• The proposed network offers a promising tool for early disease diagnosis.
• Enhanced segmentation accuracy provides clinicians with more reliable clinical insights.