Overview of the Whole Heart and Heart Chamber Segmentation Methods.

Area of Science:

• Medical Imaging
• Cardiovascular Research
• Image Processing

Background:

• Cardiovascular disease (CVD) research prioritizes heart and vessel health.
• Advanced imaging techniques enhance understanding of disease physiology and progression.
• Cardiac image processing aims for comprehensive cardiac analysis, but modality differences pose segmentation challenges.

Purpose of the Study:

• To critically review segmentation methods for the whole heart, bi-ventricles, and left atrium.
• To compare segmentation techniques across Computed Tomography (CT), Magnetic Resonance Imaging (MRI), and echocardiography (echo).
• To summarize segmentation challenges, classify contributions, and evaluate method performance and accuracy.

Main Methods:

• Methods are classified into edge-based, model-fitting, and machine/deep learning approaches.
• Segmentation methods are further categorized by the targeted cardiac structure.
• A critical review assesses the performance and accuracy of various segmentation techniques.

Main Results:

• Edge-based methods offer semi-automatic control for physicians.
• Model-fitting methods are robust to image quality variations but require prior knowledge.
• Deep learning methods show high performance with sufficient, accurate training data.

Conclusions:

• The choice of segmentation algorithm depends on imaging modality and desired level of automation.
• Each segmentation approach (edge-based, model-fitting, deep learning) has distinct advantages and limitations.
• Future research should focus on improving the robustness and data efficiency of deep learning cardiac segmentation.