A New Approach for Automatic Identification of the Ventricular Boundary in Cine Magnetic Resonance Images

Identification of the endocardial borders remains challenging in cardiology. In this paper, we propose a new approach named 'deformation flow tracking' which couples the obtained boundary from the previous frame and the extracted edges in the current frame by energy minimization. Firstly, the edges are extracted accurately by an effective threshold selection method. Then, the boundary in the previous frame is driven toward the extracted edges to form a deformation boundary by minimizing the energy between the deformation boundary and extracted edge while keeping the deformation boundary smooth. Deformation thresholds are defined and used to constrain the motions of the boundary points and eliminate outliers effectively. The proposed approach was tested on complete short-axis cine MRI datasets from 5 normal subjects and 5 patients with heart failure (total of 1660 images) randomly chosen from a much larger dataset (100 cases). As it turned out, the proposed approach is efficient and robust for automatic identification of the ventricular endocardial boundary that moves directionally and regularly, which is true in most cases