Machine learning based image segmentation for large-scale osteoarthritis analysis
Description
TitleMachine learning based image segmentation for large-scale osteoarthritis analysis
Date Created2020
Other Date2020-01 (degree)
Extent1 online resource (xiv, 84 pages) : illustrations
DescriptionOsteoarthritis (OA) is the most common degenerative joint disease worldwide, tending to occur in the joints of hip and knee. Large adult population in the United States have been affected by OA, and by 2030, an estimated 20 percent of Americans (about 70 million people) may be at increased risk for this disease. Effective medical image segmentation methods play fundamental roles in the clinical analysis of the disease. In this dissertation, three machine learning based segmentation for knee cartilage, femoral head-neck junction and thigh muscular/adipose tissue are discussed, respectively. Furthermore, large-scale OA analysis on the knee and hip joints could be further implemented based on these segments.
Knee cartilages (i.e., femoral, tibial, and patellar cartilage) are essential tissue for knee radiographic OA diagnosis. Effective segmentation of knee cartilages in large-sized and high-resolution 3D magnetic resonance (MR) data is firstly proposed. The key contribution is an adversarial learning based collaborative multi-agent network. The method employs three parallel segmentation agents to label cartilages in their respective region of interest (ROI), and then fuses the three cartilages by a ROI-fusion layer and drive a collaborative learning by an adversarial sub-network. The ROI-fusion layer not only fuses the individual cartilages, but also backpropagates the training loss from the adversarial sub-network to each agent to enable joint learning of shape and spatial constraints. The proposed scheme is shown robust and accurate in knee cartilage segmentation, and it is effective for cartilage biomarkers (e.g., surface area, volume) estimation in large-scale quantitative tests. Second, a deep multi-task learning network is exploited for the shape-preserved segmentation of the proximal part of femur (i.e., femoral head and neck) in 2D MR images. This method combines the tasks of region identification and boundary distance regression, and thus enables the task-specific feature learning for continuous segmented object with smooth boundary. This bone joint depiction could help the measurements of the femoral head-neck morphology and reflect the evolution of hip OA. In the last part of the dissertation, the muscular and adipose tissue extraction in 3D MR thigh data is investigated by an integrated framework. Specifically, deformable models and learning based detection/classification are integrated into the framework to enable robust tissue quantification for a large-scale analysis of OA-related thigh tissue changes.
NotePh.D.
NoteIncludes bibliographical references
Genretheses, ETD doctoral
LanguageEnglish
CollectionSchool of Graduate Studies Electronic Theses and Dissertations
Organization NameRutgers, The State University of New Jersey
RightsThe author owns the copyright to this work.
Statistical Profile