C. Barillot, D.R. Haynor, and P. Hellier (Eds.): MICCAI 2004, LNCS 3217, pp. 1064–1066, 2004. © Springer-Verlag Berlin Heidelberg 2004 Fig. 1. 2D Gabor filter Gabor Filter-Based Automated Strain Computation from Tagged MR Images Tushar Manglik 1 , Alexandru Cernicanu 2 , Vinay Pai 1 , Daniel Kim 1 , Ting Chen 1 , Pradnya Dugal 1 , Bharathi Batchu 1 , and Leon Axel 1 1 Department of Radiology, New York University, NY, USA 2 Department of Electrical Engineering, University of Pennsylvania, PA, USA Abstract. Myocardial tagging is a non-invasive MR imaging technique; it generates a periodic tag pattern in the magnetization that deforms with the tissue during the cardiac cycle. It can be used to assess regional myocardial function, including tissue displacement and strain. Most image analysis methods require labor-intensive tag detection and tracking. We have developed an accurate and automated method for tag detection in order to calculate strain from tagged magnetic resonance images of the heart. It detects the local spatial frequency and phase of the tags using a bank of Gabor filters with varying frequency and phase. This variation in tag frequency is then used to calculate the local myocardial strain. The method is validated using computer simulations. 1 Introduction Conventional tag analysis techniques, such as finite element and B-spline models, require tag tracking [1-2] that often rely on active contours. The purpose of this study was to develop an automated Gabor filter-based tag analysis method. Previously, Gabor filters have been used for quantifying displacement and to enhance or suppress either the tags or the non-tagged regions of the image [3-5]. 2 Methods In image domain, a set of 2D Gabor filters is used for convolution with tagged image. A two dimensional Gabor filter (Fig. 1) h(x,y) is mathematically defined as h(x,y) = g(x′,y′).s(x,y) . (1) g(x′,y′) = 1/(2л σ x′ σ y′ )exp (– ((x′/ σ x′ ) 2 + (y′/ σ y′ ) 2 )/2) . (2) s(x,y) = sin(2πd/λ+Φ), d = x cos(ξ) + y sin(ξ) (3) x’ = x cos(θ) + y sin(θ),y’ = x sin(θ) + y cos(θ) (4) where σ x′ , σ y′ are the standard deviations of the 2D Gaussian envelope along the x and the y directions, θ is the orientation of the Gaussian envelope, ξ is the orientation of the sinusoid and Φ