GPU acceleration of landmark and intensity based image registration

Abstract

Real time image registration is critical for many medical applications related to image guided navigation and functional imaging such as functional magnetic resonance imaging (fMRI), dynamic contrast-enhanced magnetic resonance imaging (DCE) and magnetic resonance spectroscopic imaging (MRSI). However, the major part of the image registration process, which is to search the spatial position given volume data, is computationally intensive. Also, to complete the image registration on a real-time basis poses high computation requirement because fast speed of computation without delay is expected. To solve the above problems, GPU is employed by utilizing its parallelizing features. In this project, the image processing techniques for salient feature extraction as well as the landmark and intensity based image registration were studied. The extraction of planes from volume data was performed by interpolation algorithm. Feature points extraction, GPU Sift Matching and Mutual Information Matching were carried out and the results were analysed and correlated with moving objects. In addition, GPU Programming Language (GLSL and OpenCL) and project development methods were learnt along the way.