Segmentation of renal compartments in DCE-MRI of human kidney

Chronic renal failure is an increasing problem world-wide. It is important to monitor renal function precisely to assess disease progression, the prognosis, and therapy-effects. Correct segmentation of kidney compartments from the DCE-MR images is needed to support automated functional analysis of kidney. In this work we compared three unsupervised data driven approaches to automatically segment the renal compartments: k-means clustering, wavelet based clustering, and Gaussian Mixture Models. The algorithms were applied to 4 DCE-MRI data sets in human kidney. Data were acquired with different temporal and spatial resolution as well as at 1.5T and 3T. Obtained results were compared to manually segmented kidney compartments. On average, the renal cortex could be segmented at 88%, the medulla at 91%, and the pelvis at 98% accuracy. Comparison of the methods showed no significant differences in the segmentation accuracies (p >;0.5). In conclusion, unsupervised data driven segmentation could be seen as a reliable and automated approach to support the analysis of renal function.