Thesis
Intercomparison of medical image segmentation algorithms
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2006
- Thesis identifier
- T11859
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- Magnetic Resonance Imaging (MRI) is one of the most widely-used high quality imaging techniques, especially for brain imaging, compared to other techniques such as computed tomography and x-rays, mainly because it possesses better soft tissue contrast resolution. There are several stages involved in analyzing an MRI image, segmentation being one of the most important. Image segmentation is essentially the process of identifying and classifying the constituent parts of an image, and is usually very complex. Unfortunately, it suffers from artefacts including noise, partial volume effects and intensity inhomogeneities.Brain, being a very complicated structure, its precise segmentation is particularly necessary to delineate the borders of anatomically distinct regions and possible tumors. Many algorithms have been proposed for image segmentation, the most important being thresholding, region growing, and clustering methods such as k-means and fuzzy c-means algorithms. The main objective of this project was to investigate a representative number of different algorithms and compare their performance. Image segmentation algorithms, including thresholding, region growing, morphological operations and fuzzy c-means were applied to a selection of simulated and real brain MRI images, and the results compared.The project was realized by developing algorithms using the popular Matlab® software package. Qualitative comparisons were performed on real and simulated brain images, while quantitative comparisons were performed on simulated brain images, using a variety of different parameters, and results tabulated. It was found that the fuzzy c-means algorithm performed better than all the other algorithms, both qualitatively and quantitatively. After comparing the performance of all algorithms, it was concluded that, by combining one or two basic algorithms, a more effective algorithm could be developed for image segmentation that is more robust to noise, considers both intensity and spatial characteristics of an image, and which is computationally efficient.
- Advisor / supervisor
- Sandham, William
- Resource Type
- DOI
Relações
Itens
Miniatura | Título | Data de carga | Acesso | Ações |
---|---|---|---|---|
|
PDF of thesis T11859 | 2023-06-14 | Público | Baixar |
![]() |
ZIP file of appendix to thesis T11859 | 2023-06-14 | Público | Baixar |