Curve Distance Measure for Quantitative Evaluation of Myocardial Boundary Tracking

Abstract

Myocardial boundary tracking is part of feature extraction task for evaluating human cardiac abnormalities. In this case, the feature of myocardial contraction is extracted in the form of the boundary shape movement and alteration such as shape motion, shape deformation.  In this paper, a curve distance measure is proposed for evaluating performance of myocardial boundary tracking techniques.  The curve is constructed of 9 – 14 control points along the myocardial boundary which is tracked to all frames in the cardiac cycle between the end diastole and the end systole. The curves distance measure implemented in this study is average of absolute shortest distance between estimated curve and ground truth curve. Both of curves are generated using b-spline interpolation technique. The estimated control point is the yield of the tracking process of the initial point in motion vector space.  In this experiment, eight seven motion estimation techniques such as Black Anandan, Brox, Gautama, Gurner Farneback, Horn Shuck, LK, LK-Pyr and proposed method are compared to know their performance in tracking on echocardiography image.  The database sets are generated to test the performance algorithm that comprising set sequences of ultrasound images with known initial and finalboundary. The known vector motion is used to generate the  ground truth of synthetic myocardial boundary tracking. The real clinical data that have initial and final boundary generated by cardiologist are also used to evaluate the performance. Accordingly, results obtained showed that quantitative measures can be obtained in term of curve distance mean of initial and final boundary which is useful for performance evaluation of the aforementioned techniques with the proposed method recorded superiority over other techniques. In short, the curve distance measure described in here is useful and provides means of evaluating performance of myocardial boundary tracking quantitatively.