The Correlation Coefficient Technique for Pattern Matching

Introduction: The ability to track motion is essential for many MRI applications. Specific applications of interest in our lab include retrospective and prospective 2D navigator echo motion artifact correction as well as post-process image averaging[l]. Previous attempts at motion tracking can be broadly classified into: 1) phase-based subtraction or 2) image-based pattern matching techniques. Pattern matching attempts to follow a region of anatomy through a series of images by finding the location in each image that best matches a pre-selected pattern containing the anatomy of interest. An advantage of pattern matching is that it is less sensitive to spurious phase that may arise from a variety of sources in MRI. Additionally, even if non-rigid body motion is present (e.g. in the heart), pattern matching may still be used to follow a local region of the anatomy. The major disadvantage of current pattern matching techniques is that they frequently fail if there are large interor intra-image intensity variations. Such a situation is encountered, for example, with flowing blood in the heart. In this abstract, the correlation-coefficient pattern matching technique is discussed. This technique is completely insensitive to image intensity variations. Additionally, it can be performed extremely rapidly, thus making it useful for techniques such as prospective navigation. Theory: Previous pattern matching techniques have used least squares[2] (LS) and/or cross-correlation[l,2] (XC) as the matching procedure. The LS technique finds the location (to, ~0)) in the image, that minimizes the least-square distance between the object (f(z, y)) and the MxN pattern (h(z, y)):