Please use this identifier to cite or link to this item:
|Title:||A medical image-based computer-aided diagnosis system for musculoskeletal disease and disorder.||Authors:||Chuah, Tong Kuan||Keywords:||DRNTU::Engineering::Bioengineering||Issue Date:||2012||Source:||Chuah, T. K. (2012). A medical image-based computer-aided diagnosis system for musculoskeletal disease and disorder. Doctoral thesis, Nanyang Technological University, Singapore.||Abstract:||Musculoskeletal diseases such as osteoarthritis (OA) are responsible for a large number of disabilities among the world’s population. With the increase of aging population and increase in the amount and complexity of medical data that clinicians need to handle, computer-aided diagnosis (CAD) system is becoming increasingly important in improving the efficiency, reproducibility and accuracy of the diagnosis process. CAD systems for breast cancer detection are currently being used in clinical practice but the use of CAD for musculoskeletal diseases is still under research. This thesis makes several advancements in the research and development of an image-based CAD system for musculoskeletal diseases, mainly focusing on OA and various aspects of the CAD system. The CAD system is designed to provide supplementary information to support medical decision, or to provide second opinion to the practitioner in cases of ambiguity. An image-based CAD system typically has components such as a segmentation (or feature extraction) module, a measurement module, a classification module and/or a visualization module. This thesis made advancements in these components. Cartilage defect is an important biomarker for OA. It is important to be able to visualize damaged cartilage during diagnosis using medical images to ascertain the size and locations of the defects. To aid the visualization of damaged cartilage, the thesis first developed a visualization framework for visualizing cartilage damage or lesion in proton density weighted MR images. Using the cartilage visualization framework developed, it is possible to effectively display damaged cartilage. A metric has also been studied for its ability to correlate with the percentage of damaged cartilage. A linear relationship between percentage of damaged cartilage and the metric studied was found. As part of the advancement to classifying subjects with BML, the thesis investigated textural parameters as potential biomarker in separating between bone marrow with and without bone marrow lesion. Through statistical analysis, a set of parameters was identified and was further used for classification of image slices and subjects so that a second opinion could be provided by the computer. The classification results confirmed that image textural information of bone marrow can provide reasonably accurate results in differentiating between subjects with and without BML: the area under receiver operating characteristic (ROC) curve achieved is 0.914. Having established the ability to classify subjects with and without BML, the thesis then deals with the development of an automated segmentation algorithm for the bone in knee MRI, by proposing a new termination criterion and an initialization strategy for active contours. Segmentation is an important and necessary step before features can be quantified to be used in the analysis and classification system. For automatic implementation of active contours in which different shapes need to be segmented, the proposed termination criterion demonstrated almost 50% and 60% total time reduction while achieving similar accuracy as compared with conventional pixel movement-based method in the segmentation of synthetic and real medical images, respectively. The initialization strategy worked as expected and achieved DSC of 96.7 ± 1.1% for the data validated. Overall, the thesis has made a balanced advancement to various components of the CAD system for musculoskeletal diseases, forming the foundation for future work to incorporate more anatomical structures of the joints (ligaments, muscles etc.) and biomarkers in the diagnosis, and further improving the segmentation process.||URI:||https://hdl.handle.net/10356/52048||DOI:||10.32657/10356/52048||Fulltext Permission:||open||Fulltext Availability:||With Fulltext|
|Appears in Collections:||SCBE Theses|
Page view(s) 20449
Updated on Jul 27, 2021
Updated on Jul 27, 2021
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.