Jump to: navigation, search

Cardiac Material Markers from Tagged MR Images

Williams S. Kerwin and Jerry L. Prince



We have developed a technique for tracking the motion of the left ventricle of the heart during contraction.  The technique uses a sequence of magnetic resonance images and an established imaging method known as "tagging".  We use tagged MR data in a unique method that produces 3-D motion estimates over a discrete set of points (so-called material points) in the wall of the left ventricle.  The method relies on a multi-step process that includes use of thin-plate splines to interpolate data between image planes and an iterative, alternating projections algorithm for locating the tracked material points.  The details of the method are presented below. 

Image of material pointsAn example of the resulting data is presented at the left (click on the image to view an mpeg movie).  The movie shows the position of some 254 material points at ten time frames during the contraction of the left ventricle.  The points are viewed down the center of the left ventricle.  Watch for the characteristic twisting of the left ventricle.  Such data can be used for visualization of heart function or for developing models of heart motion.  We are also working to extract clinically useful information for the early detection and assessment of ischemic heart disease. 




Measurement of heart wall motion during contraction is a fundamental challenge in the study of cardiac mechanics.  Motion measurement has the potential to be used as a diagnostic tool for assessment of heart disease and as a theoretical tool for physiological analysis.  Past approaches for measuring heart motion can be divided into two types. One type tracks the movement of a set of points - called material markers - which are either implanted or naturally occuring landmarks within the heart.  The second type of approach uses tagged magnetic resonance imaging techniques to estimate the displacement field for all points within the heart.  Here we present a technique that uses the standard MR tagging protocol, but measures motion at a discrete set of material points as in the material marker methods.  We refer to the resulting motion data as "MR markers".  This technique may be used to compare the merits of MR tagging to material marker techniques, or to provide data to existing analysis routines intended for use with material marker data.


Tagged Heart ImageWe use standard tagged MRimages, where "tagging" is a method that aids in tracking motion of objects using a series of magnetic resonance images.  Tags appear as dark regions in the images that move with an object as it moves.  Tagging is particularly valuable in cardiac imaging because the tissue in the walls of the heart provides few natural features for motion tracking.  An example of a tagged heart image is shown at left, where the ring shaped object is the wall of the left ventricle.  The tag pattern used was originally a set of parallel stripes and this image was taken after heart motion had deformed the stripes into the bent lines you see here.  Click on the image to see a 10 frame mpeg movie demonstrating the deformation of the tag lines from their original straight configuration.

Although "tag lines" is the normal term for these induced features, tags generally consist of sheets of dark (when imaged) tissue. Tag lines are the intersection of these "tag surfaces" with image planes.  This concept is illustrated in the next image below, which shows a single tag surface intersecting a set of images.  The tag surface contributes one tag line to each of the image planes.  A standard tagged MRI data set consists of many such tag surfaces arranged in a regular 3-D grid.  The contracting heart deforms this regular grid and from such deformation, motion information is extracted.  To illustrate this point, click on the second image below to view a movie showing the deforming grid.

Tag Surface Intersecting Image Planes Deformed Grid of Tag Surfaces
Although this geometry of tag surfaces has been used by other researchers (see for example the Medical Imaging Lab of The Johns Hopkins University School of Medicine), we recognized that within this regular grid, there are unique points whose 3-D motion can be determined exactly from the motion of the grid.  Specifically, the nodes of the grid - those points where any three surfaces intersect - are tied to specific points within the heart tissue.  Thus, for these points and only these points, the 3-D motion can be determined solely on the basis of the deforming grid.  Therefore, our method tracks all grid intersections within the wall of the left ventricle.

Our complete method is a three step process consisting of grid reconstruction, surface intersection identification, and segmentation.  Grid reconstruction must be performed because the position of the tag surfaces is only known within image planes.  Between image planes, the tag surfaces are estimated using thin-plate splines, which individually interpolate each surface.  The complete reconstructed grid (as depicted above) is the combination of the thin-plate spline surfaces.  Next, all grid intersection points are identified using an iterative, alternating-projections algorithm.  In the publications listed, we offer a proof that this algorithm converges to the unique intersection point of any three surfaces.  The identified points are then the nodes of a regular 3-D grid as depicted in the next image.  Finally, the points are segmented to identify only those within the wall of the left ventricle.  The result is a cup shaped set of MR markers as shown in the second image below.

Image of Grid Intersection Points Segmented Intersection Points


This method has been applied to several data sets.  Here we present a typical example.  The tagged MRI protocol was applied to a healthy human volunteer.  The initial distance between tag surfaces was 6mm and images were taken every 32.5msec for ten times during contraction of the heart.  After applying our algorithm, 254 MR markers were identified within the wall of the left ventricle.  Images of the results are depicted below.  The first image is a stereo pair viewing the left ventricle down the long axis (from base toward apex).  This image can be viewed in 3-D by focusing your left eye on the left image and right eye on the right image.  A 3-D movie is also available by clicking on the image.  Viewing this image in 3-D is useful for demonstrating the distribution of the markers in space.  Finally, the last image is a side view of the MR markers with the free wall of the left ventricle on the left.  Click on the image to view a movie version of this image, which will demonstrate the vertical shortening of the left ventricle during contraction.

 Stereo Pair View of MR Markers
Side View of MR Markers


  • W.S. Kerwin and J.L. Prince, ``Generating 3-D Cardiac Material Markers Using Tagged MRI,'' Proc. Information Processing in Medical Imaging (IPMI), Berlin: Springer-Verlag, pp. 313-326, June 1997. [request hard copy]
  • W. S. Kerwin and J. L. Prince, "Recursive Filtering and Interpolation Algorithm for Function Sequences of Any Dimension", Proc. Image and Multidim. Dig. Sig. Proc., pp.239-242, Alpbach, July 1998.
  • W. S. Kerwin and J. L. Prince, "MR Tag Surface Tracking Using a Spatio-temporal Filter/Interpolator", Proceedings of IEEE Int. Conf. Image Proc., vol. 1, pp. 699-703, Chicago, October 1998.
  • W. S. Kerwin and J. L. Prince, "Cardiac Material Markers from Tagged MR Images," Medical Image Analysis, vol. 2, issue 4, pp. 339-353, 1998.
  • W. S. Kerwin and J. L. Prince, "Tracking MR tag surfaces using a spatio-temporal filter and interpolator," Int'l Journal of Imaging Systems and Technology, vol. 10, pp. 128-142, 1999.
  • W. S. Kerwin and J. L. Prince, "The Kriging Update Model and Recursive Space-Time Function Estimation," IEEE Trans. on Signal Processing, vol. 47, no. 11, pp. 2942-2952, November 1999.
  • W. S. Kerwin and J. L. Prince, "Kriging Filters for Space Time Interpolation," Advances in Imaging and Electron Physics, invited by Dr. Peter Hawkes, 124:139-193, November 2002.