Tracking Endocardium Using Optical Flow along Iso-Value Curve

In cardiac image analysis, optical flow techniques are widely used to track ventricular borders as well as estimate myocardial motion fields. The optical flow computation is typically performed in Cartesian coordinates, and not constrained from a priori knowledge of normal myocardium deformation patterns. However, for cardiac motion analysis, displacements along specific directions and their derivatives are usually more interesting than 2D or 3D displacement fields themselves. In this context, we propose a general frame work on optical flow estimation along iso-value curves. We applied the proposed framework in a specific application: for endocardium tracking on cine cardiac MRI series. The endocardial surfaces tracked with the proposed algorithm were quantitatively compared with manual tracing at each frame. The proposed method was also compared to the regular Lucas-Kanade optical flow method directly applied to MRI image data in Cartesian coordinates. Quantitative comparison showed a positive improvement in average tracking errors, through the whole cardiac cycle