3D active appearance models: application to cardiac MR and ultrasound image segmentation

This paper presents a novel, 3D model-based method for three-dimensional image segmentation: A 3-D Active Appearance Model (AAM) is reported as an extension of the 2D AAM framework introduced by Cootes and Taylor. The model's behavior is learned from manually-traced segmentation examples during an automated training stage. Information about shape and image appearance of the cardiac structures is contained in a single model. This ensures a spatially- and/or temporally consistent segmentation of three-dimensional cardiac images. The clinical potential of the 3-D Active Appearance Model is demonstrated in short-axis cardiac magnetic resonance (MR) images and four-chamber echocardiographic sequences. The method's performance was assessed by comparison with manually-identified independent standards in 56 clinical MR and 64 clinical echo image sequences. The AAM method showed good agreement with the independent standard using quantitative indices of border positioning errors, endo- and epicardial volumes, and left ventricular mass. The 3D AAM shows high promise for successful application to MR and echocardiographic image analysis in a clinical setting.