Shape model segmentation of long-axis contrast enhanced echocardiography

We segment long-axis, four-chamber contrast enhanced echocardiography imagery using active shape models. The active shape model algorithm uses principal component analysis to model the shape variability found in a database of training shapes. Accurate segmentation from this model is accomplished by applying a specialized gradient vector flow field to guide the contours to the myocardial borders. The success of the proposed algorithm was verified by application to 65 patient myocardial contrast echocardiography (MCE) studies and through comparison with manually drawn contours. This approach improved accuracy over previously reported data, providing an average accuracy of 0.98, sensitivity of 0.84, specificity of 0.99, and RMSE of 3.3 pixels, when compared to ground truth. Error and variability from automatic segmentation were found to be less than those among multiple human observers. The approach also compares favorably against a standard active contour solution