Cardiac image segmentation from cine cardiac MRI using graph cuts and shape priors

Area of Science:

• Medical Imaging
• Computer Vision
• Cardiology

Background:

• Accurate segmentation of cardiac structures, specifically the left ventricle, right ventricle, and myocardium, is crucial for diagnosing heart conditions using cine cardiac magnetic resonance (CMR) images.
• Existing segmentation methods often struggle with poor edge definition and significant shape variations within and between patients, necessitating improved techniques.

Purpose of the Study:

• To develop and evaluate a novel segmentation method for cardiac structures in CMR images.
• To incorporate prior shape information within a graph cut framework to enhance segmentation accuracy, addressing limitations of existing approaches.

Main Methods:

• A two-stage segmentation approach is proposed, utilizing a graph cut framework.
• The method incorporates prior shape information derived from distance functions and orientation angle histograms, adapted for individual datasets due to interpatient variability.
• A two-stage process involves initial segmentation using intensity information, followed by a second stage combining intensity and shape priors to refine results.

Main Results:

• Experimental results on 30 real patient datasets demonstrate higher segmentation accuracy when shape information is utilized.
• The proposed method shows superior performance compared to other competing segmentation techniques.
• The approach effectively handles challenges posed by poor edge information and significant shape variations.

Conclusions:

• The novel graph cut-based method effectively segments cardiac structures in CMR images by integrating prior shape information.
• The strategy of adapting shape priors to individual datasets addresses interpatient variability, leading to improved accuracy.
• This method offers a robust and accurate solution for cardiac MRI segmentation, outperforming existing approaches.