A Computer-Aided Diagnostic System for the Early Detection of Prostate Cancer Using Diffusion-Weighted Magnetic Resonance Imaging

Abstract

The area of medical image analysis has always been an inspiring application of the various computer vision and pattern recognition theories and approaches. Due to its challenging aspects, prostate cancer diagnosis via magnetic resonance images (MRI) is one of the most popular in the medical imaging research community. Several studies have highlighted the need to develop a robust and accurate computer-aided diagnosis (CAD) system to detect and localize suspicious lesions in the prostate area. Accordingly, this thesis addresses the topic of MRI-based prostate cancer detection and localization. We begin with an extended review of the state-of-the-art CAD systems, identifying the gaps and limitations of the di?erent approaches. Afterward, we propose our two contributions for 1) Computer-aided design system or prostate cancer detection; and 2) Automatic prostate cancer tumor detection. In the first contribution, we employ a classical pattern recognition and machine learning approach to classify benign and malignant subjects. In this system, we utilize Diffusion-weighted (DW) images at different magnetic field strengths to produce a global apparent diffusion coefficient (ADC) map for each subject. These maps are then reshaped and classified using a shallow neural network. In the second contribution, we propose a deep convolutional encoder-decoder architecture to segment tumors in T2-weighted (T2W) images. We also provide a multi-channel approach to exploit the 3D information in the MR images without compromising in the computational cost.

Date of Award	Dec 2018
Original language	American English