Automatic segmentation of left ventricle endocardium from cardiac MR images using active contours driven by local and global intensity fitting energy

Abstract

In this paper, we present a fully automated method for segmenting left ventricle endocardium from multi slice cine short axis cardiac MR images. Our method does not require manually drawn initial contour and is able to segment images in the presence of noise and intensity inhomogeneity. The segmentation process flow uses temporal variance of image intensity to localise the heart region. Slices are segmented sequentially using a local and global statistics-based active contour model. To control the influence of the global energy, an adaptive weight function that varies dynamically with image region is applied. The method was tested on a database of 30 cases obtained from the Sunnybrook Health Sciences Centre, and the results were compared with manual delineated ground truth. The algorithm’s performance is evaluated using two metrics, average perpendicular distance (APD) and dice similarity coefficient (DSC). Resulting contours show a mean DSC of 0.88 and an overall APD around 2 mm. Linear regression analysis of ejection fraction (EF) yielded a slope 1.015 and R2 = 0.926. The proposed segmentation approach shows a better performance and provides a practical method for use in clinical practice.