Coronary artery segmentation in CCTA images based on multi-scale feature learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease Research

Background:

• Coronary artery segmentation is crucial for diagnosing Coronary Artery Disease (CAD).
• Current segmentation methods for Coronary Computed Tomography Angiography (CCTA) images face challenges like manual intervention and low accuracy.
• Existing approaches struggle to effectively address these segmentation difficulties.

Purpose of the Study:

• To propose a novel Multi-scale Feature Learning and Rectification (MFLR) network.
• To achieve automatic and accurate segmentation of coronary arteries.
• To overcome the limitations of existing coronary artery segmentation techniques.

Main Methods:

• The MFLR network utilizes a multi-scale feature extraction module in the encoder for capturing diverse contextual information.
• A feature correction and fusion module in the decoder uses high-level features to refine low-level features.
• This module fuses features across levels to enhance segmentation performance.

Main Results:

• The MFLR network demonstrated superior performance across key metrics including Dice similarity coefficient, Jaccard index, Recall, F1-score, and 95% Hausdorff distance.
• These results were consistent on both in-house and public datasets.
• The network achieved the best performance among evaluated methods.

Conclusions:

• The MFLR approach exhibits superiority and strong generalization capabilities in coronary artery segmentation.
• This advancement contributes to more accurate diagnosis and treatment of Coronary Artery Disease.
• The findings have implications for other medical image segmentation applications.