Design and implementation of parallel algorithms for medical imaging applications

The objective of this research is to design and implement fast algorithms for detecting blood vessels in infrared images of laser-heated skin. Information about location, depth, diameter, and orientation of blood vessels is important for proper selection of laser treatment parameters. The input image is generated by a high speed infrared focal plane array (IR-FPA) camera and consists of a time sequence of recorded infrared emission images.; In this research, a three-dimensional Tomographic Reconstruction Algorithm (3D TRA) has been developed and implemented on a parallel machine at the Maui High Performance Computing Center (MHPCC). Execution of the 3D TRA is computationally intensive. To reduce the execution time, the serial computation of a 3D TRA is decomposed into multiple tasks which can be executed in parallel. 3D TRA is a technique which calculates the 3D image of initial space-dependent temperature increase from the time sequence of recorded infrared emission images.; The tomographic image contains information about blood vessels located beneath the skin's surface. Algorithms for blood vessel detection based on 3D invariant moments are designed. These algorithms accept the output image from the parallel 3D TRA and calculate location, length, diameter, and orientation of detected blood vessels. The algorithms are tested on the infrared images provided by Beckman Laser Institute at UCI, and experimental results are presented in this study.; 3D reconstruction of laser heated discrete subsurface chromophores in human skin recorded by infrared radiometry is an important clinical problem for improved laser treatment. This research presents algorithms which may be used to compute wavelength of laser radiation used for treatment. The main purpose of this dissertation is to employ enhanced computational tools for better and improved laser treatment of vascular disorders (e.g., Port Wine Stain).