Automated left ventricular cardiac MRI segmentation, quantification and visualization using a radial-ray approach

Abstract

A new segmentation algorithm is proposed that automatically segments the endocardium, epicardium and papillary muscles of the left ventricle (LV) from short-axis cardiac MRI data sets. The algorithm is based on image intensity gradient analysis using higher order derivatives and local parameterisation to obtain reliable segmentation results. User interaction is minimised to just the definition of centre points for the ventricle base, apex and papillary muscles. From the approximate centre of the ventricle on each slice, rays are cast and then scanned to find the extent of the ventricle wall. The segmentation algorithm parameterises three dynamic intersection tests to detect endocardium and epicardium intersections for each emitted ray. A high-resolution mesh describing both surfaces of the myocardium is then constructed from these intersections. Automated segmentation and reconstruction of the papillary muscles allows their exclusion from LV volume calculations and their inclusion in myocardial mass calculations. A software package was developed to evaluate the algorithm, providing DICOM compatibility and user interaction for segmentation corrections. Three-dimensional visualization of the segmentation results is also catered for, as both still and animated meshes, in mono or stereo vision. Currently, the algorithm detects endocardium intersections with great accuracy and the algorithm is robust even on problematic data sets. It also provides accurate reconstruction of the papillary muscles.