Abstract:
This study presents a novel method for building parametric representations of myocardial microstructure of the left ventricle from multi-directional diffusion weighted magnetic resonance images (DWI). The direction of maximal diffusion is directly estimated from the DWI signal intensities using finite element field fitting. This framework avoids the need to compute diffusion tensors, which introduces errors due to least squares fitting that are generally neglected when building microstructural models of the heart from DWI. Nodal parameters describing cardiac myocyte orientations throughout a finite element model of the left ventricle were fitted to a series of raw diffusion signals using non-linear least squares optimisation to determine the direction of maximum diffusion. An ex vivo DWI data set from a Wystar-Kyoto rat was processed using the proposed method. The fitted myocyte orientations were compared against conventional diffusion tensor/eigenvector analysis and the degree of correlation was measured using a normalised dot product (nDP). Good agreement (nDP=0.979) between the new method and the traditional tensor analysis approach was observed for regions of high fractional anisotropy (FA). In regions of low FA, the errors were much more variable, but the proposed method maintains a smoothly varying myocyte angle distribution as is generally used in tissue and organ scale heart models.