Date on Master's Thesis/Doctoral Dissertation


Document Type

Doctoral Dissertation

Degree Name

Ph. D.


Computer Engineering and Computer Science

Degree Program

Computer Science and Engineering, PhD

Committee Chair

Elmaghraby, Adel S.

Committee Co-Chair (if applicable)

El-Baz, Ayman

Committee Member

Nasraoui, Olfa

Committee Member

Imam, Ibrahim

Committee Member

Beache, Garth


Prostate--Cancer--Diagnosis; Diagnostic imaging


Prostate cancer is the second most fatal cancer experienced by American males. The average American male has a 16.15% chance of developing prostate cancer, which is 8.38% higher than lung cancer, the second most likely cancer. The current in-vitro techniques that are based on analyzing a patients blood and urine have several limitations concerning their accuracy. In addition, the prostate Specific Antigen (PSA) blood-based test, has a high chance of false positive diagnosis, ranging from 28%-58%. Yet, biopsy remains the gold standard for the assessment of prostate cancer, but only as the last resort because of its invasive nature, high cost, and potential morbidity rates. The major limitation of the relatively small needle biopsy samples is the higher possibility of producing false positive diagnosis. Moreover, the visual inspection system (e.g., Gleason grading system) is not quantitative technique and different observers may classify a sample differently, leading to discrepancies in the diagnosis. As reported in the literature that the early detection of prostate cancer is a crucial step for decreasing prostate cancer related deaths. Thus, there is an urgent need for developing objective, non-invasive image based technology for early detection of prostate cancer. The objective of this dissertation is to develop a computer vision methodology, later translated into a clinically usable software tool, which can improve sensitivity and specificity of early prostate cancer diagnosis based on the well-known hypothesis that malignant tumors are will connected with the blood vessels than the benign tumors. Therefore, using either Diffusion Weighted Magnetic Resonance imaging (DW-MRI) or Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI), we will be able to interrelate the amount of blood in the detected prostate tumors by estimating either the Apparent Diffusion Coefficient (ADC) in the prostate with the malignancy of the prostate tumor or perfusion parameters. We intend to validate this hypothesis by demonstrating that automatic segmentation of the prostate from either DW-MRI or DCE-MRI after handling its local motion, provides discriminatory features for early prostate cancer diagnosis. The proposed CAD system consists of three majors components, the first two of which constitute new research contributions to a challenging computer vision problem. The three main components are: (1) A novel Shape-based segmentation approach to segment the prostate from either low contrast DW-MRI or DCE-MRI data; (2) A novel iso-contours-based non-rigid registration approach to ensure that we have voxel-on-voxel matches of all data which may be more difficult due to gross patient motion, transmitted respiratory effects, and intrinsic and transmitted pulsatile effects; and (3) Probabilistic models for the estimated diffusion and perfusion features for both malignant and benign tumors. Our results showed a 98% classification accuracy using Leave-One-Subject-Out (LOSO) approach based on the estimated ADC for 30 patients (12 patients diagnosed as malignant; 18 diagnosed as benign). These results show the promise of the proposed image-based diagnostic technique as a supplement to current technologies for diagnosing prostate cancer.