3D bilateral filtering of cardiac DT-MRI data

In this paper, the 3D bilateral filtering (3DBF) method of cardiac diffusion tensor magnetic resonance imaging (DT-MRI) data is proposed. The technique integrates information both coming from spatial localization of the diffusion tensors (DTs) as well as angular similarity (AS) of primary eigenvectors of DTs and Jensen-Bregman LogDet (JBLD) divergence for covariance matrices. In comparison to recently literature reports, we show that using a similarity measure based solely on a covariance matrices is not enough to perform the most accurate DT-MRI data enhancement in tensors domain. Our investigation on synthetic and real ex-vivo canine cardiac DT-MRI data shows that the best results by 3DBF method are commonly obtained by AS measure and JBLD divergence synergy. The fractional anisotropy root mean-squared error (FA RMSE) for cardiac data decreased from 0.0855 (noisy data) to 0.0473, whereas primary eigenvectors angle difference mean (AD) improved from 25.9±19.8 to 19.2 ± 18.9.