Quantification of Biomechanical Imaging Biomarkers
PhD-defence by Sudhakar Tummala
Osteoarthritis (OA) is a debilitating musculoskeletal disorder in the elderly and also a major burden for healthcare economy in western countries. Biomechanics may play a vital role in the early stages of OA. The dissertation presents refinements of exiting method for quantification of Cartilage Surface Smoothness (CSS), and novel methods to quantify Contact Area (CA) and Congruity Index (CI) in the Medial Tibio-Femoral (MTF) cartilage compartment non-invasively using low-field magnetic resonance imaging (MRI). Initially, to reduce the voxellation effects, the MTF binary cartilage compartments were regularized using mean curvature flow in a level-set formulation before quantifying CSS, CA and CI. The first and second order Gaussian derivatives of the signed distance representation of level-set surfaces were computed and used for the computation of the CSS as inverse of local mean curvature.
The CA was quantified by employing the voxel width as threshold. The local CI was quantified in the CA by assessing first and second order general surface features and associating them. The quantifications were validated on a longitudinal study population from the greater Copenhagen area. The CA was significantly different between healthy and early OA subjects. The CI was also able to separate healthy from early OA.
The CSS and CI performed better than MTF volume for diagnosis of early radiographic OA. Further, the CSS, CA and CI showed suitable as efficacy markers by demonstrating strong correlations with percentage longitudinal changes with MTF cartilage volume. Therefore, the CSS, CA and CI may be the denominators in the OA progression. The automatic segmentations also in general allowed the CSS/CS/CI performed equally or better than those from manual segmentations for diagnosis. Female MTF joints demonstrated greater normalized CAs and lower CIs. The lower CIs may help to explain the more prevalence of female OA. Future work could focus on validation of these markers on larger study populations.
Chairman: Associate Professor Jon Sporring, Department of Computer Science, Copenhagen University
Member 1: Professor Sebastien Ourselin, Deputy Director, Centre for Medical Image Computing Reader in Medical Image Computing University College London
Member 2: Professor Sharmila Mujamdar, Department of Radiology and Biomedical Imaging, UCSF Professor Dept. of Orthopedic Surgery,UCSF and Professor, Dept. of Bioengineering, UC Berkeley
Academic supervisor: Professor Mads Nielsen, Department of Computer Science, Copenhagen University