TAG SEPARATION IN CARDIAC TAGGED MRI J. HUANG Presenter: Lin Zhong MICCAI 2008 MRI Magnetic resonance imaging(MRI) primarily a noninvasive medical imaging technique used in radiology to visualize detailed internal structure and limited function of the body MRI provides much greater contrast between the different soft tissues of the body than computed tomography (CT) TAGGED MRI Tagging consists in tattooing the myocardium with a geometrical pattern (lines or grid), using selective spatial pulses. the magnetization of the myocardial (or any other) tissue will persist for times (e.g. one cardiac circle) and will create a visible mark in the MR images that moves precisely with the underlying tissue. can noninvasively render visible the internal motions of tissues. Pose great challenges to the cardiac image processing RELATED WORK Morphological operations [Guttman 1994] Fill in the region between removed tagging lines Bad generalization Band pass [osman 1999] Enhance tag pattern region and increase contrast Performance depend on the designed filters. RELATED WORK Band stop [Qian,Z. 2007] Images are processed in the spectral domain Low frequency peak at the origin is from actual tissue Other energy peaks are introduced by tag patterns Abandon all frequency in the localized regions will cause some artifacts in the recovered images after tag removal MOTIVATION The tag patterns have a regular texture Discrete Cosine Transform (DCT) The Cardiac images without tag patterns are piecewise smooth with sparse gradients Wavelet Transform (WT) Two dictionaries can be built. Tag pattern and piecewise smooth image can be represented sparsely by these two dictionaries. Tag-only image can be used for tag tracking, the remaining image can be used for accurately localizing the cardiac boundaries. PROBLEM FORMULATION I=S+T I : tagged cardiac MR image S: piecewise smooth cardiac image without tags T: tag-only image PROBLEMS TO BE SOLVED Build dictionary A for cardiac images S Undecimated Wavelet Transforms (UWT) Suitable for the sparse representation of natural scene Build dictionary B for tag pattern T Discrete Cosine Transform (DCT) Deal images in the frequency domain, and is appropriate for a sparse representation of periodic or smooth behaviors. Find optimal coefficients for Image decomposition via the combination of sparse representations and a variational approach. ITIP, 2005 OPTIMIZATION OF COEFFICIENTS Original Objective function Modified objective function TV penalty guide S to be a piece wise smooth image with sparse gradient. Block-coordinate Relaxation algorithm[Bruce,1998] RESULTS