Search results
(1 - 1 of 1)
- Title
- IMAGE PROCESSING ALGORITHMS FOR PROSTATE CANCER LOCALIZATION WITH MULTISPECTRAL MAGNETIC RESONANCE IMAGING
- Creator
- Xin, Liu
- Date
- 2011-11, 2011-12
- Description
-
In this thesis, we develop a series of image processing algorithms to localize prostate cancer with multispectral magnetic resonance (MR)...
Show moreIn this thesis, we develop a series of image processing algorithms to localize prostate cancer with multispectral magnetic resonance (MR) images to guide biopsy, surgery and minimally invasive therapy. Besides, we develop a new method to for evaluation of image classification algorithms considering correlation between neighboring pixels. Prostate cancer is one of the most prevalent cancer types and one of the leading causes of cancer death among men in the United States. High-resolution MRI has shown higher accuracy than trans-rectal ultrasound (TRUS) to ascertain the presence of prostate cancer. In this work, three different types of MR techniques are employed to provide both morphological and functional information about the benign and malignant tissues of the prostate. These are T2-weighted (T2w) MRI, diffusionweighted imaging MRI (DWI) and dynamic contrasted enhanced MRI (DCE MRI). In the first chapter of this thesis, we briefly describe the fundamentals of different MR techniques, and the multispectral MR dataset used in our experiment. Then, we focus on two tasks of the prostate cancer localization problem: prostate gland extraction and prostate tumor localization. For each topic, we review the previous studies available in the literature, and present our methods with their advantages. Finally, the new image evaluation method considering correlation between pixels is presented. Our prostate segmentation method is fully unsupervised and extracts the prostate gland from DWI MRI in 3D by fusing the active contour model and shape prior information. For tumor localization, we develop an unsupervised approach which is based on fuzzy Markov random field (fuzzy MRF) model, a new scheme based on relative intensity values which can be combined with supervised segmentation classifiers to mimic the cancer localization procedures performed by human readers and a new feature named location map which incorporates the spatial inforx mation of the tumors to remove the need for manual peripheral zone extraction. The proposed image evaluation algorithm is based on receiver operating characteristics (ROC) analysis and it considers the correlation between neighboring pixels. This method could replace the conventional ROC analysis and offers a more accurate evaluation of the test image. Our algorithms are tested on 20 patients’ multispectral MR images, and the qualitative as well as quantitative experimental results demonstrate the efficacy of our segmentation methods and show that the proposed segmentation methods outperform the currently available used approaches. The evaluation method has been tested on computer simulated images and shows very promising results. The summary and future work is also described at the end of the thesis.
Ph.D. in Electrical Engineering, December 2011
Show less