Dilated convolution network with edge fusion block and directional feature maps for cardiac MRI segmentation

Area of Science:

• Medical imaging analysis
• Cardiovascular imaging
• Deep learning in healthcare

Background:

• Accurate cardiac magnetic resonance imaging (MRI) segmentation is crucial for assessing cardiac function.
• Existing deep learning methods struggle with fuzzy boundaries and class ambiguity in cardiac MRI.
• Limited utilization of multi-scale features and inadequate boundary information hinder segmentation performance.

Purpose of the Study:

• To develop an advanced deep learning model for precise cardiac MRI segmentation.
• To address the challenges of fuzzy boundaries and class ambiguity in cardiac MRI segmentation.
• To improve the evaluation of cardiac function through enhanced segmentation accuracy.

Main Methods:

• A dilated convolution network incorporating an edge fusion block and directional feature maps was proposed.
• Multi-scale features were fused using dilated convolutional layers during downsampling.
• An edge fusion block integrated boundary information, and a direction field module refined segmentation features.

Main Results:

• The proposed method demonstrated superior performance on the automated cardiac diagnosis challenge (ACDC) and myocardial pathological segmentation (MyoPS) datasets.
• Comparative experiments confirmed the effectiveness of the novel network architecture.
• The approach successfully addressed limitations of existing methods in utilizing multi-scale and boundary information.

Conclusions:

• The developed cardiac MRI segmentation method offers improved accuracy and robustness.
• The integration of edge fusion and directional features enhances segmentation quality.
• This technique provides a more reliable tool for cardiac function evaluation using MRI.