PET technology has an important role in modern medical diagnostics. With this process we can view snapshot of a given part of the body's metabolism which provides more information than examining the organ's anatomy. Therefore illnesses can be detected earlier, chance for recovery increases.
Depending on the examined metabolic process and organ, a given type of molecule is marked with radioisotope which emits positrons under its decomposition. The positrons collide with the body's electrons and become gamma photons which impact with the detectors after escaping the body. From photon-pairs detected by the detectors we can compute the radioisotope's concentration distribution which we can use to conclude the examined organ's metabolism. A measurement could contain billions of photon-pairs which we use one by one in case of list mode reconstruction, thus we are able to assign the full measurement data to each of them, contrary to the ordinary approach where we build our computations upon the statistics of the properties of the photon-pairs. The drawback of the list mode approach is that we have to process much more data, the reconstruction time is evidently longer. We solve this computationally intensive but well paralellizable problem using graphics card (GPU) which has the necessary computing capacity and parallel architecture. In addition the list mode reconstruction demands other optimalization considerations because the algorithm's input data has different structure. Despite the drawbacks of the list mode approach it will take an important part of the PET technology as the GPU technology develops.
The thesis' purpose is to explain the list mode reconstruction system's theory and the challenges of its implementation on the GPU, implement the list mode algorithm and compare it with the existing algorithm, analyze its efficiency and quality of the results.
The results are encouraging, the list mode approach can compete with the ordinary one, however, there are lots of improvements to be made.