Real-time imaging and automated segmentation of cartilage from 2D ultrasound images
Description
TitleReal-time imaging and automated segmentation of cartilage from 2D ultrasound images
Date Created2018
Other Date2018-10 (degree)
Extent1 online resource (xi, 71 pages) : illustrations
DescriptionKnee osteoarthritis (OA), a chronic joint condition, occurs when the cartilage cushion between the knee joints degrades over the age. The progression of OA is determined based on the cartilage degradation, therefore, cartilage thickness is an important measure for diagnosis and classication of OA. Currently, magnetic resonance imaging (MRI) is used as the gold standard imaging modality for the diagnosis of OA. However, the routine clinical monitoring using MRI is limited as MRI is expensive, has high scanning time, and limited accessibility. Recently, ultrasound (US) has shown its sensitivity to evaluate the cartilage changes. US provides cost-effective, and real time imaging of knee joint thus making it a potential alternative to MRI for routine clinical monitoring of cartilage degeneration. However, low contrast, speckle noise, signal attenuation, shadow artifacts, and being a user dependent imaging modality have prohibited the widespread use of this imaging modality for diagnosing OA. Various studies have been conducted to show the potential of US for routine clinical monitoring of OA. However, the studies were only focused on qualitative assessment of US cartilage image and the cartilage thickness was computed manually. This thesis, presents a fully automated cartilage segmentation and thickness computation from enhanced 2D US knee cartilage images.
The proposed framework consists of: (1) cartilage image enhancement, (2) bone surface segmentation for seed initialization, (3) seed-based cartilage segmentation, and (4) automatic cartilage thickness measurement. Local phase image features, by designing various bandpass quadrature lters, are extracted for enhancing the cartilage image, and bone surfaces. The segmentation of enhanced bone surfaces is achieved using a dynamic programming approach. The extracted bone surfaces are marked as an initial seeds for region based segmentation algorithms. Cartilage segmentation is evaluated using random walker (RW), watershed, and graph-cut methods. During the final step the segmented cartilage regions are used to compute mean cartilage thickness. The qualitative and quantitative validations are performed on 200 2D scans obtained from ten healthy volunteers. Validation against expert manual segmentation achieved mean dice similarity coecient (DSC) of 0.90, 0.86, and 0.84 for RW, watershed, and graph-cut respectively. The computed mean cartilage thickness ranged from 1.90 to 5.66 mm with the average value of 2.95 0.66mm. The Bland-Altman plots are used to compare the mean thickness error among dierent type of segmentation algorithm. This study presents a, fully automated US cartilage segmentation approach for cartilage segmentation. Presented results show the potential of US for imaging cartilage. The work will be invaluable for all future studies investigating US as an alternative imaging modality in OA research with a specic focus on cartilage.
NoteM.S.
NoteIncludes bibliographical references
Genretheses, ETD graduate
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.