Enhancement of human cardiac DT-MRI data using locally adaptive filtering

A main source of problem in human cardiac diffusion tensor magnetic resonance imaging (DT-MRI) is the presence of high noise that buries useful information such as edges, details, and even whole objects. We propose a new enhancement approach based on the combined use of non-stationarity degree (NSD) notion and amoebas algorithm. The method consists of first calculating the NSD at each pixel of the diffusion weighted (DW) image, then introducing the NSD in the amoebas formulation, and finally determining the regions using the thus modified amoebas, inside which the pixels are averaged. The results obtained on human cardiac DW images show that edges and details in the DW images are almost preserved while the homogeneous regions are well smoothed, and that the resulting principle eigenvector field and fiber bundles are more coherent.