Atalas Construction for Cardiac structure and motion

    To achieve the goal, it is necessary to develop algorithms for cardiac motion estimation and co-registration across different individuals. In this project, the computational processes of estimation of cardiac motion within individuals and deformable registration of motion across individuals will be coupled into a unifying framework of 4-dimensional (4D) deformable registration that simultaneously estimates cardiac motion within individuals and detects correspondences across individual cardiac image sequences, thereby enabling accurate construction of a statistical atlas of cardiac structure and function for reliably distinguishing subtle manifestations of disease from normal variation.In the following, the current results on cardiac atlas construction, cardiac motion estimation, and myocardial fiber orientation estimation are provided.

• Construction of a cardiac atlas at end-systole [1]

    Spatial normalization of individual data to a predefined template is a necessary step for constructing a cardiac atlas. It increases the statistical power for analysis of differences between normal and pathologic states, and also allows insights into structural or functional changes from the transformation maps that describe the variations of the individuals from the atlas. We applied our HAMMER registration algorithm to short axis cine GRASS MR images of the hearts obtained at end-systole in 20 normal volunteers (mean age 39 years, range 21 to 79), and created a cardiac atlas by averaging the 20 spatially normalized datasets. As shown in Fig 1, the average image is relatively clear, which partly indicates the accuracy of our registration algorithm.

• Consistent cardiac motion estimation [2]

    We have extended 4D-HAMMER for consistent registration of sequences of volumetric MR images to an atlas. We required the deformation fields to be spatially smooth for registering the neighboring frames in the sequence, and also temporally smooth for consistent motion estimation. Promising results for consistent motion estimation were achieved and are demonstrated in Fig 2.

    We validated our method by comparing our estimated motion with that obtained using tagged MR data [50, 51]. Both tagged and untagged MR images were acquired from healthy volunteers on the same scanner during the same scanning session at end-expiration, thus the datasets were assumed self-registered. To quantize the accuracy of motion estimates, tag-intersection points were manually detected and tracked by an observer in three different SA slices and one LA slice over all time frames between end-diastole and end-systole, including 20 points on the SA slices and 14 points on the LA slices. The root mean square error between the in-plane displacements estimated by our method and the actual in-plane displacements as measured by the observer was within acceptable limits (2mm).

    In order to further evaluate the accuracy of the motion estimation obtained using our method, we extracted the motion fields from the tagged MR images using the method proposed by Chandrashekara et al [26]. We compared the determinant of the Jacobian, calculated in blocks corresponding to the tag spacing (8 mm). The average error of the determinant of the Jacobian between the two methods was 5.63% with a standard deviation of 3.44%.

• Myocardial fiber orientation estimation [3]

    Myocardial fiber orientations are important for accurate modeling of cardiac electromechanics. However, it is extremely difficult to estimate these directly in vivo with current imaging techniques. Most cardiac models in use currently use synthetic models for fiber orientations which fail to capture subtle variations in fiber orientations. We present a way to map the diffusion tensors from a template onto patient specific cardiac geometry, using elastic registration followed by a reorientation of the diffusion tensors based on the local rotation component of the transformation. The effectiveness of the diffusion tensor mapping is validated on a set of diffusion tensor datasets obtained from 19 canine subjects. The algorithm was able to map the diffusion tensors effectively for both healthy and failing hearts. Fig. 3 shows an example of constructed myocardial fibers in a heart.

Reference:

1. Harold Litt, Dinggang Shen, and Christos Davatzikos, "Registration and Spatial Normalization of Cardiac MR Images Using HAMMER", SCMR (Society for Cardiovascular Magnetic Resonance) 2004 Conference. (Also in Journal of Cardiovascular Magnetic Resonance, Vol 6(1), 2004 as Abstract.)
2. Dinggang Shen, Hari Sundar, Zhong Xue, Yong Fan, Harold Litt, "Consistent Estimation of Cardiac Motions by 4D Image Registration", MICCAI, Palm Springs, California, USA, Oct 26~29, 2005.
3. Hari Sundar, Dinggang Shen, George Biros, Harold Litt, Christos Davatzikos, "Estimating Myocardial Fiber Orientations by Template Warping", Third IEEE International Symposium on Biomedical Imaging (ISBI 2006), April 6-9, 2006, Arlington, VA, USA.