Reconstruction of Patient-Specific Left Atrial Geometry from CMR Imaging

Abstract

Clinical quantification of left atrial (LA) volumes is currently performed using either the biplane area-length method or the method of discs (Simpson‘s method) on 2-D cardiac images or 3D echocardiography. However, these methods tend to underestimate LA volumes as compared to cardiovascular magnetic resonance (CMR) imaging. In this paper, we propose a geometry-based reconstruction algorithm for computing the LA volume automatically for the entire cardiac cycle by combining information from both the short- and long-axis from CMR imaging. The inputs to our reconstruction algorithm are as follows: (i) a set of segmented short-axis contours and (ii) a set of segmented long-axis contours from the standard 2-chamber and 4-chamber views. Our approach consists of a series of iterative steps where the most basal short-axis contour is projected in the atrial direction and subsequently morph to the patient-specific LA shape using the long-axis contours as guide. These series of morphing generate a left heart comprising both the LV and LA geometries with a planar basal surface. To reconstruct the LA cap, this planar basal surface is morphed into a hemisphere representing the closed surface of the LA using the long-axis contours as guide, thereby allowing us to reconstruct the closed LA shape and to calculate its volume.