Automatic renal segmentation applied in pediatric MR Urography

Abstract

In the field of pediatric MR Urography research, renal segmentation is an important task which is tedious and error prone when performed manually. An automatic renal segmentation method is proposed in this paper which is aimed at providing researcher with a useful tool for MR Urography. The method accomplishes the segmentation with set of clustering and inferring modules. In the first module, the features in the temporal dimension are described by 5 principal components and their spatial relationships are modeled by the Markov random field. On the assumption of the finite mixture model, the whole 4D data are grouped into clusters by an expectation maximization algorithm. In the second module, using a priori anatomical and functional knowledge, a two-stage recognition is performed to label each cluster first into kidney and background, then into the different compartments including cortex, medulla and collection system. The automatic results are tested using actual clinical data sets and validated by comparison with the manual results. The comparisons demonstrate good agreements in both the anatomical and functional analysis.