Enhancement of myocardial boundary tracking using wavelet-based motion estimation

Myocardial boundary tracking in echocardiograms is a challenging task due to soft tissue contrast, speckled noise, scattering and attenuation of the ultrasound signal. Furthermore, most ultrasound images that are acquired by physicians in clinical practice have a poor rating quality and are hard to analyze and recognize. Both of these factors could complicate the development of an algorithm to track the movement of myocardial boundary in echocardiograms. With this in mind, we proposed a method that combines a wavelet multi-scale strategy and a warping optical flow to generate a high-accuracy velocity vector from two consecutive frames of poor-quality ultrasound images. From these sets of high-accuracy velocity vectors, the movement of points along the myocardial boundary is tracked starting from the end diastole to the end systole of the cardiac cycle. A set of multi-scale images generated by Haar wavelet decomposition is processed recursively to compute the motion vector field in an echocardiographic image sequence. Artificially generated cardiac image sequences were used to measure performance by comparing the angular error of the proposed motion estimation technique to other established methods. The proposed method was also tested and evaluated by expert cardiologists using actual poor-quality ultrasound images that were acquired from healthy and unhealthy volunteers to track myocardial boundaries based on the parasternal long axis view of the human cardiac.