Medical imaging techniques such as CT, MRI and x-ray imaging are a crucial component of modern diagnostics and treatment. As a result, many automated methods involving digital image processing have been developed for the medical field. Image segmentation is the process of finding the boundaries of one or more objects or regions of interest in an image. This thesis focuses on accelerating image segmentation for the localization of cancerous lung nodules in two-dimensional radiographs. This process is used during radiation treatment, to minimize radiation exposure to healthy tissue. The variational level set method is used to segment out the lung nodules. This method represents an evolving segmentation boundary as the zero level set of a function on a two-dimensional grid. The calculus of variations is employed to minimize a set of energy equations and find the nodule's boundary. Although this approach is flexible, it comes at significant computational cost, and is not able to run in real time on a general purpose workstation. Modern graphics processing units offer a high performance platform for accelerating the variational level set method, which, in its simplest sense, consists of a large number of parallel computations over a grid. NVIDIA's CUDA framework for general purpose computation on GPUs was used in conjunction with three different NVIDIA GPUs to reduce processing time by 11x--20x. This speedup was sufficient to allow real-time segmentation at moderate cost.
Library of Congress Subject Headings
Image processing--Digital techniques; Classification--Data processing; Computer graphics; Imaging systems in medicine
Department, Program, or Center
Computer Engineering (KGCOE)
Prosser, Nathan, "Medical image segmentation using GPU-accelerated variational level set methods" (2010). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus