A major problem in cardiac imaging is the measurement of cardiac motion for identification of ischemic tissues. Magnetic resonance (MR) tagging has demonstrated enormous potential for clinical use since tagged images reveal motion within the myocardial tissue, not just at the boundaries. In past work tagged images have revealed evidence of both torsional motion of the left ventricle (LV) and regional thickening of the LV myocardium. Tagging methods have also been used to measure a spatial pattern of 3-D strain tensors in the myocardium around the LV chamber. All previous tagged image analysis methods, however, have focused on matching extracted features within image pairs, which sacrifices resolution and permits gross errors from erroneous matches. In contrast, optical flow (OF) analysis promises to measure myocardial motion pixel- by-pixel with high accuracy and no gross errors. This grant application proposes to further develop optical flow methods for automated 3-D motion analysis of left ventricular motion. Specifically, we propose to 1) develop improvements to variable brightness optical flow (VBOF), a new optical flow technical developed by the principal investigator for MR tagging; 2) implement multiscale optical flow (MS)F) and develop new multiscale optimal tag patterns; 3) develop 3-D methods, and 4) develop a complex motion simulator. Improvements of VBOF will concentrate on reducing its sensitivity to brightness variations and improving its particle tracking methods. Multiscale optical flow methods will be developed primarily to decrease the computational time of the proposed 3-D methods. Previous work has shown that tag patterns which optimize OF performance occupy a lower frequency band than planar tags. We propose to develop fast imaging methods that selectively scan k-space in the low frequency band, thus imaging the pattern with high quality and low noise. Since VBOF will be jointly optimized for this pattern, fast imaging should actually improve out motion analysis. Since understanding the 3-D motion of the heart is vital, we will explore two ways to implement OF in 3-D. The first, to extend our current methods to 3-D, is readily accomplished; an additional constraint based on the assumption that the myocardium is nearly incompressible will be added to reduce ill- conditionedness. The second approach is based on the use of 1-D tag patterns imaged in orthogonal planes, a new an very promising idea. Both approaches will be tested in simulations, phantoms, and humans. A new complex motion simulator will be developed to permit studies having known realistic (complex) motions. In summary, this grant application proposes a significant departure from the conventional MR tagging methods which use high-frequency tag patterns and feature-matching methods. With low- frequency tag patterns and OF methods, our approach has the potential to automatically produce cardiac motion estimates in 3-D with both high accuracy and high resolution.
