Home
Resource
Towards robust and effective shape prior modeling
PDF
PDF format is widely accepted and good for printing.
Plug-in required
PDF-1(14.08 MB)
Citation & Export
View Usage Statistics
Staff View

Citation & Export
Hide

Simple citation

Zhang, Shaoting. Towards robust and effective shape prior modeling. Retrieved from https://doi.org/doi:10.7282/T33N22DJ

Export

Click here for information about Citation Management Tools at Rutgers.

Statistics
Hide

Description

TitleTowards robust and effective shape prior modeling
NameZhang, Shaoting (author); Metaxas, Dimitris N (chair); Pavlovic, Vladimir (internal member); Eliassi-Rad, Tina (internal member); Kikinis, Ron (outside member); Rutgers University; Graduate School - New Brunswick
Date Created2012
Other Date2012-01 (degree)
SubjectComputer Science, Shapes, Computer simulation, Shapes--Computer simulation
Extentxvii, 99 p. : ill.
DescriptionOrgan shape plays an important role in many clinical practices, including diagnosis, surgical planning and treatment evaluation. It is usually derived from medical images using low level appearance cues. However, due to diseases and imaging artifacts, low level appearance cues are often weak or misleading. In this situation, shape priors become critical to infer and refine the shape derived from image appearances. Effective modeling of shape priors is challenging because: 1) shape variations are complex and cannot always be modeled by parametric probability distributions; 2) a shape instance derived from image appearance cues (called an input shape) may have significant errors; and 3) local details of an input shape may be important for clinical purposes but difficult to preserve if they are not statistically significant in the training data. In this paper we propose a novel Sparse Shape Composition model (SSC) to address these three challenges in a unified framework. With our method, a sparse set of shapes is selected from the shape repository and composed together to infer and refine an input shape. This way, the prior information is implicitly incorporated on-the-fly. Our model leverages two sparsity observations of the input shape instance: 1) the input shape can be approximately represented by a sparse linear combination of shapes in the shape repository; 2) parts of the input shape may contain large errors but such errors are sparse. Our model is formulated as a sparse learning problem. Using $L1$ norm relaxation, it can be solved by an efficient expectation-maximization (EM) framework. Furthermore, this model is extended to effectively handle multi-resolution, local shape priors and hierarchical priors. We also propose a framework to generate high quality training data in 3D. Our framework includes geometry processing methods and shape registration algorithms. The proposed shape prior model is extensively validated on five different medical applications: 2D lung localization in chest X-ray images, 3D liver segmentation in low-dose Computed Tomography (CT) scans, 3D segmentation of multiple rodent brain structures in Magnetic Resonance (MR) microscope, real time tracking of left ventricles in Magnetic Resonance Imaging (MRI), and high resolution CT reconstruction. Compared to state-of-the-art methods, our model exhibits better performance in all these studies.
NotePh.D.
NoteIncludes bibliographical references
NoteIncludes vita
Noteby Shaoting Zhang
Genretheses, ETD doctoral
Persistent URLhttps://doi.org/doi:10.7282/T33N22DJ
Languageeng
CollectionGraduate School - New Brunswick Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Version 8.5.5
Rutgers University Libraries - Copyright ©2024