In this paper we present a novel tag removal method for tagged cardiac MR images. This method is based on a 2D bandstop filtering technique, which selectively attenuates the tag pattern’s frequency components in the spatial-frequency domain. To automatically design the band-stop filter, we developed a mean shift-based clustering method to partition the power spectrum image into several harmonic regions. Then a principal component analysis-based multivariate fitting procedure is implemented to model the harmonic peak. We have tested this tag removal method on a variety of tagged cardiac MR images with different imaging settings, which turned to be robust and computationally modest. Comparing with previous tag removal methods [1, 2], the results of this new method are more consistent with the original image’s appearance, and can recover more detailed structures, such as the papillary muscles. Based on the new tag-removed images, we have tested segmenting the left ventricular endo...
Zhen Qian, Rui Huang, Dimitris N. Metaxas, Leon Ax