Three dimensional Time Resolved Segmentation of the Left Ventricle

Background: Segmentation of the left ventricle (LV) is of great interest since it directly gives important clinical parameters such as end-diastolic volume (EDV), end-systolic volumes (ESV), ejection fraction (EF), left ventricular mass (LVM) etc. Earlier methods for segmenting the left ventricle have used a wide range of techniques such as level sets [1], probabilistic models [2], active appearance models [3]. Active appearance and probabilistic models have problems with data sets that lie outside the learning material, and level set methods have difficulty including enough a priori information to make them robust. Method: We propose a segmentation algorithm based on the concept of deformable models, but extended with an enhanced and fast edge detection scheme that includes temporal information and anatomical a priori information. The algorithm takes a stack of time resolved short axis images and performs the segmentation simultaneously within the time-resolved 3D volume, in order to utilize as much of the available image data as possible. The model is initialized as a ‘cone ’ in the left ventricle and is allowed to expand and deform towards the endocardial surface. In the deformable model the used forces are an inflation balloon force, an edge force, a curvature force, a slice force, damping and acceleration forces, and a user interaction force. To enable wall motion studies the papillary muscles are excluded. This is achieved after a coarse segmentation and fitting of an approximately circular shape to the extracted contour. The final position of the contour is subsequently found after a few additional iterations using the deformable model. The mass/volume of the papillary muscles is excluded from the LV volume calculations, but included in the LVM calculation. This is done by detecting the papillary muscles within the blood pool using an automatically calculated threshold. The epicardial segmentation i