Meeting Banner
Abstract #1318

Renal Compartment Segmentation by Wavelet-Based Clustering of 3D DCE-MRI of Human Kidney

Frank G. Zoellner1, 2, Sheng Li1, 3, Andreas D. Merrem1, Jarle Roervik, 24, Arvid Lundervold, 45, Lothar R. Schad1

1Computer Assisted Clinical Medicine, Heidelberg University, Mannheim, Germany; 2Section for Radiology, Dept. of Surgical Sciences, University of Bergen, Bergen, Norway; 3Institute of Knowledge Based Engineering, School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, China; 4Dept. of Radiology, Haukeland University Hospital, Bergen, Norway; 5Dept. of Biomedicine, University of Bergen, Bergen, Norway

Correct determination (segmentation) of the renal compartments within the images is crucial to obtain i.e. whole kidney GFR via pharmacokinetic modelling. We propose a wavelet-based segmentation method to group the voxel time courses and thereby segment the renal compartments. This method was applied to DCE-MRI data sets of 4 healthy volunteers and 3 patients. On average, the renal cortex could be segmented at 88%, the medulla at 91%, and the pelvis at 98% accuracy. Time intensity curves showed well known characteristics of perfusion time curves for the respective compartments. In conclusion, wavelet based clustering of DCE-MRI of kidney is feasible.

Abstract PDF Poster PDF

Keywords

accuracy acquisition applied artifacts assessment assisted assoc benign bergen biomedicine characteristic china clinical cluster clustering clusters coefficients compartment compartments computation computer consists corrected cortex cosine costly courses crucial curves cyst degree delay delineations depending depicted depicting dept described description detected detection diagnosed diameter disease diseases distance early elsewhere engineering estimator evaluation evolved examination examined example explained failure female filtration findings frank function future gained give good graph healthy human illustrate important input institute insufficiency intensity kidney kidneys known lava life like localized long male manual masses matrix mechanical medulla morphology mother motion none nonparametric part partition patient patients patterns pelvis people perfusion plot precise prevent procedure progression promising propose radiology recorded renal respective risk sampling scanner school section segment segmentation segmented sets seven shanghai slice space sparse spatial speed stein steps sufficiently support therapies thereby third thresholded thresholding tong towards transplantation treatment typical unbiased vibe volunteer volunteers wavelet whereas whole years