Regional perception and multi-scale feature fusion network for cardiac segmentation

Area of Science:

• Medical imaging analysis
• Cardiovascular disease diagnosis
• Artificial intelligence in healthcare

Background:

• Cardiac magnetic resonance (CMR) imaging is crucial for cardiovascular disease (CVD) assessment.
• Accurate segmentation of cardiac structures in CMR images is challenging due to heart motion and similar tissue grayscale values.
• Precise segmentation is vital for quantifying cardiac function and aiding clinical decisions.

Purpose of the Study:

• To develop a more accurate segmentation approach for CMR images.
• To enhance the precision of cardiac structure segmentation for improved CVD assessment.
• To reduce the burden on radiologists through automated segmentation.

Main Methods:

• Proposed a regional perception and multi-scale feature fusion network (RMFNet) for CMR image segmentation.
• Introduced window selection transformer (WST) and grid extraction transformer (GET) modules for enhanced feature perception.
• Incorporated a novel multi-scale feature extraction module to preserve detailed information.

Main Results:

• RMFNet demonstrated superior performance compared to advanced methods on three cardiac datasets.
• The model achieved excellent generalizability, outperforming comparison methods on a multi-organ dataset.
• Experimental validation confirmed the effectiveness and robustness of the RMFNet approach.

Conclusions:

• RMFNet offers a significant advancement in CMR image segmentation accuracy.
• The developed method can improve the efficiency and reliability of CVD diagnosis.
• Accurate segmentation plays a key role in supporting radiologists and guiding clinical interventions.