Vollständige iterative Rekonstruktion von dreidimensionalen Positronen-Emissions-Tomogrammen unter Einsatz einer speicherresidenten Systemmatrix auf Single- und Multiprozessor-Systemen

摘要

Positron emission tomography (PET) is a modern medical diagnostic procedure, which enables non invasive views of the metabolism of living organisms. Thereby it visualises also malfunctions which are characteristic for neurological, cardiological and oncological illnesses. For this purpose, radioactive, positron emitting tracers are injected into the patient. The emitted radiation is measured by detector systems and images of the activity distribution are calculated by reconstruction procedures. Newer, high sensitive PET devices with 3D detector systems pose substantial, so far yet unaccomplishable demands to the reconstruction programs because of the complexity and size of data. In this work a procedure is presented, with which the reconstruction problem is represented as linear equation system (LES). Methods are developed and implemented, in order to calculate the coefficient matrix of the LES exactly as well as efficiently and to compress it so extensively, that it can be stored completely in memory. Here the LES can be solved by means of a fast converging, iterative approximation method in short computation time. In addition, the well-established Ordered-Subset-MLEM algorithm is parallelized, the subset partitioning is improved, and the handling sequence of the subsets is optimized. This method of reconstruction is superior to known procedures regarding image quality and computation time.