Conical deformable model for myocardial segmentation in late-enhanced MRI

This paper presents a conical 3D deformable template for fully automatic and robust segmentation of late-enhanced MRI (LE-MRI) datasets. The proposed technique has several advantages over existing works. Firstly, it uses a thick-walled conical model that is suitable to derive fully automatic and reliable initialization by taking into account potential short-axis misalignments. Furthermore, it uses to its advantage the geometrical and appearance properties of the blood pool to decouple the endocardial and epicardial segmentations. The final epicardial result is obtained using thickness smoothness measures constrained on the initial robust segmentation of the endocardium. Detailed validation using 20 LE-MRI datasets and comparison to previous work demonstrates that the technique is reliable and promising for clinical assessment of LE-MRI data.