This thesis develops novel algorithms to detect early acute rejection of cardiac transplants imaged by magnetic resonance (MR) protocols.
The thesis integrates tagged MRI data and diffusion-tensor MRI (DT-MRI) data to derive cardiac 3D motion and 3D strain maps. The deformation of tag lines in tagged MR images assists in tracking the cardiac motion. However, due to limited imaging time, tagged MRI data is spatially and temporally undersampled, which makes it very difficult to track the 3D heart voxels through the cardiac cycle. To overcome the spatial undersampling, we model the heart muscle by a fibrous architecture; and, to address the temporal undersampling, we describe the fibers elongation and contraction with a biomechanics motion model. We fit the fibrous architecture to the in vivo heart by estimating the fiber orientations with diffusion tensor MR imaging of a fixed heart. We reconstruct the 3D motion and strain maps of the heart through the cardiac cycle by minimizing an energy functional. This functional couples the internal energy of the biomechanics fibrous model with an external energy term that tracks the constancy of the voxel intensities across frames.
To localize the myocardial regions undergoing rejection, we develop an automatic classifier to determine the abnormal motions and strain. We use a graph to capture the strain similarities among myocardial voxels in the 3D strain map. Then, we design a Cheeger constant based algorithm to partition the graph into two disjoint subgraphs, one representing the healthy regions and the other representing the dysfunctional regions. We also apply the automatic classifier on ultrasmall superparamagnetic iron oxide (USPIO) enhanced MRI data to monitor the immune cell infiltration.
The experiments with real MRI data of rejecting rat allografts demonstrate the good performance of the algorithms developed in this thesis. We evaluate the 3D motion estimator by comparing its results with manual tracking; the validation shows that the estimator has small errors, 0.3 pixels average deviating from the ground truth. We compare the performance of our automatic classifier with manual classification by an expert; the performance evaluation concludes that our classifier agrees with the expert classification on 96% of the voxels, 6% to 18% higher than the performance of several other standard classifiers.
|Advisor:||Moura, Jose Manuel Fonseca de|
|School:||Carnegie Mellon University|
|School Location:||United States -- Pennsylvania|
|Source:||DAI-B 68/09, Dissertation Abstracts International|
|Subjects:||Biomedical research, Electrical engineering|
|Keywords:||Cardiac transplants, Heart rejection, Strain maps|
Copyright in each Dissertation and Thesis is retained by the author. All Rights Reserved
dissertation or thesis. The supplemental files are provided "AS IS" without warranty. ProQuest is not responsible for the
content, format or impact on the supplemental file(s) on our system. in some cases, the file type may be unknown or
may be a .exe file. We recommend caution as you open such files.
supplemental files is subject to the ProQuest Terms and Conditions of use.