SEGMENTATION OF MYOCARDIUM USING DEFORMABLE REGIONS AND GRAPH CUTS

Area of Science:

• Medical Imaging
• Computational Anatomy
• Cardiovascular Imaging

Background:

• Cardiac MRI is crucial for Left Ventricle (LV) assessment.
• Accurate segmentation of LV epicardium and endocardium is challenging.
• Existing segmentation methods like deformable models and graph cuts have limitations.

Purpose of the Study:

• To develop a novel, semi-automatic segmentation system for LV epicardium and endocardium in MRI.
• To combine the strengths of deformable models and graph cuts for improved segmentation.
• To enhance accuracy, smoothness, and reduce interaction costs in cardiac segmentation.

Main Methods:

• A hybrid approach coupling deformable models with graph cuts was developed.
• Temporal information from consecutive cardiac phases was exploited.
• A deformable model defined by two nested contours was used for unified segmentation.
• A single energy functional was optimized for simultaneous epicardium and endocardium segmentation.

Main Results:

• The coupled system provided accurate cues for graph cuts and good initialization for deformable models.
• The unified contour approach ensured inherent coherency between epicardium and endocardium.
• Achieved more accurate and smoother segmentation results compared to using graph cuts alone.
• Demonstrated reduced interaction costs in segmentation tasks.

Conclusions:

• The proposed system effectively segments the Left Ventricle epicardium and endocardium in MRI.
• Combining deformable models and graph cuts offers synergistic benefits for cardiac image segmentation.
• The unified, nested contour approach enhances segmentation accuracy and consistency.