Left ventricle segmentation in MRI via convex relaxed distribution matching

Area of Science:

• Medical Imaging
• Cardiovascular Diseases
• Computational Anatomy

Background:

• Accurate segmentation of the left ventricle (LV) in cardiac magnetic resonance images (MRIs) is crucial for diagnosing cardiovascular diseases.
• Existing automatic segmentation algorithms often require extensive training data or significant user interaction, limiting their clinical applicability.
• Challenges include handling shape variations and achieving computational efficiency for 3D cardiac MRI volumes.

Purpose of the Study:

• To develop a fast and efficient algorithm for automatic detection of left ventricle (LV) endocardium and epicardium surfaces in cardiac MRIs.
• To reduce the need for large training datasets and intensive user input in cardiac MRI segmentation.
• To improve the speed and accuracy of LV segmentation using convex relaxation and distribution matching techniques.

Main Methods:

• The proposed algorithm utilizes convex relaxation and distribution matching, incorporating shape and intensity priors for LV segmentation.
• It requires minimal user input (a single point per region) and training from a single subject, achieving scale-invariance.
• The method employs a parallelized approach using a graphics processing unit (GPU) and the augmented Lagrangian method to solve optimization problems.

Main Results:

• The algorithm achieves segmentation in approximately 3.87 seconds per cardiac MRI volume, representing a five-fold speed-up compared to standard implementations.
• Performance evaluation over 400 volumes demonstrates that the generated 3D surfaces correlate well with manual delineations.
• The algorithm's performance is robust and not significantly affected by the choice of the training subject.

Conclusions:

• The developed algorithm offers a fast, accurate, and user-friendly solution for left ventricle segmentation in cardiac MRIs.
• Its ability to train on a single subject and its computational efficiency make it suitable for clinical applications.
• The method effectively handles shape deformations and reduces reliance on costly registration procedures and large training sets.