STACS: new active contour scheme for cardiac MR image segmentation

Area of Science:

• Medical Imaging
• Image Segmentation
• Computational Anatomy

Background:

• Cardiac magnetic resonance (MR) imaging presents segmentation challenges due to low contrast, artifacts like papillary muscles, and turbulent blood flow.
• Accurate segmentation of cardiac structures is crucial for quantitative analysis and diagnosis in cardiovascular medicine.

Purpose of the Study:

• To develop and evaluate a novel stochastic active contour scheme (STACS) for robust and automatic segmentation of cardiac MR images.
• To overcome specific challenges in cardiac MR image segmentation, including low contrast, papillary muscles, and turbulent blood flow.

Main Methods:

• STACS minimizes a combined energy functional incorporating stochastic region-based and edge-based information, shape priors, and local contour properties.
• A level set method is employed to solve the Euler-Lagrange equation governing contour evolution.
• An annealing schedule dynamically balances energy functional terms, enhancing robustness and segmentation accuracy.

Main Results:

• STACS successfully segmented cardiac structures (ventricles, epicardium) from surrounding tissues (chest wall) in 48 real cardiac MR images.
• The method demonstrated robustness to noise and initial contour placement due to integrated edge and region-based information.
• Segmentation accuracy was validated against manual tracings using area and edge similarity measures, showing excellent performance.

Conclusions:

• The stochastic active contour scheme (STACS) provides a powerful and accurate tool for automatic cardiac MR image segmentation.
• STACS effectively handles common challenges in cardiac MR imaging, offering a reliable solution for clinical applications.
• The proposed method achieves high segmentation consistency and performance, comparable to manual segmentation standards.