Skip to Main Content
We introduce a method for the heterogeneity analysis of liver tissue and other enhanced organs in the Dynamic contrast enhanced magnetic resonance imaging (DCE-MRI) of the liver based on the similarity of enhancement patterns of signal curves. This analysis is done by an iterative piecewise hierarchical clustering method. The novel idea of the clustering algorithm is a new similarity measure which compares the wash out part of the signal curves for parallelism. To enhance signal-to-noise ratio, the signal curves are derived from a reduced subspace with the help of principle component analysis. The method is evaluated on nine DCE-MRI liver datasets from patients with different kinds of tumors. The heterogeneity analysis is based on the fact that the distance between the signal curves of the voxels belonging to a homogeneous group to the mean curve of this group is smaller than the distance of the signal curves to the mean of a heterogeneous group with the same number of voxels. On the other hand, as the liver is a heterogeneous organ, there are voxels inside the tissue, which neighbors are part of several homogeneous areas. The experiments on nine liver datasets show that depending on the expected granularity, the tissue of liver can consist of many sub regions. This method is a suitable way to increase signal to noise ratio. The clustering gives us information about how many significant different signal curves are present in a dataset. This information helps the estimation of number of compartments and the complexity of the system.