Extraction of epi-cardium contours from unseen images using a shape database

Accurate segmentation of the myocardium in cardiac magnetic resonance images can be restricted by image noise and low discrimination between the epi-cardium boundary and other organs. Segmentation of the epi-cardium is important for the calculation of left ventricle mass. In this paper we propose a novel method of epi-cardium segmentation, which firstly segments the left ventricle cavity. The epi-cardium boundary is found using the edge information in the image, and where such information is lacking it enhances the shape with the best fitting scaled segment, taken from a database of expertly assisted hand segmented images. In the final stage the segments are connected using a natural closed spline. The method was evaluated using a leave-one-out strategy on 24 volumes and calculates the coefficient of determination as 0.93 and a root mean square of the point to curve error of 1.54 mm when compared to manually segmented images.