» master's thesis abstract
Gradient-Oriented Boundary Profiles for Shape Analysis Using Medial Features
Gradient-oriented boundary profiles have been developed as a novel method to parameterize boundaries. Boundary profiles are created at locations of high gradient magnitude by averaging intensity within a neighborhood of voxels oriented along the image gradient, making them rotationally invariant and relatively insensitive to image noise. A cumulative Gaussian is fit to the collection of averaged voxel intensities yielding estimates of (1) extrapolated intensity values for voxels located far inside and outside of a boundary and (2) anatomical boundary location. Intrinsic measures of confidence have been developed to eliminate low-confidence parameter estimates. Thresholds placed on these measures of confidence allow for high-confidence unsupervised classification of boundaries.
The validity of gradient-oriented profiles is demonstrated on artificially generated three-dimensional test data and shown to accurately parameterize and classify the boundary. Applying the measures of confidence and establishing thresholds, the accuracy of boundary location and intensities estimates improved drastically, making them a high-quality replacement for simpler methods of boundary detection. Towards shape analysis, gradient-oriented boundary profiles are applied to an existing a medial-based approach to shape analysis, known as core atoms. Core atoms in their previous implementation were based on simple gradient direction and unable to form without a priori knowledge of object intensity relative to background. Boundary profiles were applied to core atoms permitting the formation of so called "core profiles". Core profiles remove any restriction on the object's or the background's intensity, allowing multiple objects of differing intensities to be located with a single application.
Core profiles were applied to 3D computer-generated data, as well as RT3D ultrasound cardiac phantom data. It was shown on computer-generated data that calculating the volume with core profiles is more accurate then calculating the volume with core atoms, because of the improved accuracy of the boundary location. Two new methods of automatically measuring volume on non-parametric data with core profiles are proposed. Future work with includes constructing medial node models improved by gradient-oriented boundary profiles for automated left ventricular identification and measurement.
Legal Stuff: I hereby certify that, if appropriate, I have obtained and attached hereto a written permission statement from the owner(s) of each third party copyrighted matter to be included in my thesis, dissertation, or project report, allowing distribution as specified below. I certify that the version I submitted is the same as that approved by my advisory committee. I hereby grant to University of Pittsburgh or its agents the non-exclusive license to archive and make accessible, under the conditions specified below, my thesis, dissertation, or project report in whole or in part in all forms of media, now or hereafter known. I retain all other ownership rights to the copyright of the thesis, dissertation or project report. I also retain the right to use in future works (such as articles or books) all or part of this thesis, dissertation, or project report.
