Graph cut-based method for segmenting the left ventricle from MRI or echocardiographic images

Area of Science:

• Medical Imaging
• Computational Anatomy
• Cardiovascular Imaging

Background:

• Accurate 3D segmentation of the left ventricle (LV) endocardial wall is crucial for cardiac function assessment.
• Existing segmentation methods face challenges with low-contrast, noisy echocardiography and detailed, anisotropic MRI data.

Purpose of the Study:

• To present a fast, interactive 3D graph cut segmentation method for the LV endocardial wall.
• To adapt the method for both magnetic resonance imaging (MRI) and echocardiography modalities.
• To leverage a 3D Bezier coordinate system for improved segmentation accuracy and topological guarantees.

Main Methods:

• Developed a 4-step segmentation process utilizing a 3D Bezier coordinate system.
• Warped 3D images into Bezier space for implicit shape priors and hole-free results.
• Employed a max-flow min-cut algorithm on an energy graph incorporating image gradients and anatomical constraints.
• Applied post-processing convex hull algorithm for surface refinement.

Main Results:

• Achieved segmentation times of 2-5 seconds per 3D volume.
• Demonstrated superior performance on the CETUS echocardiographic dataset compared to state-of-the-art methods.
• Showed results statistically equivalent to human operators on MRI datasets.

Conclusions:

• The Bezier coordinate system-based graph cut method offers a fast and accurate solution for 3D LV endocardial segmentation.
• The approach effectively handles the distinct characteristics of MRI and echocardiography.
• This method provides a robust and efficient tool for clinical cardiovascular image analysis.