Beta 1


Title Parallel Algorithms for Ultrasound Beamformation
Author Gustavsson, Sara
Supervisor Jensen, Jørgen Arendt (Biomedical Engineering, Department of Electrical Engineering, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Nikolov, Svetoslav (Biomedical Engineering, Department of Electrical Engineering, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2002
Abstract The purpose of this master project was to develop parallel algorithms for the off-line beamformation of data collected by the experimental ultrasound system RASMUS. The result is the MATLAB toolbox called Beamformation Toolbox II. The toolbox consists of a set of user functions, which can be combined in various ways to support different techniques for beamformation, such as synthetic aperture imaging in two and three dimensions. The calculations can be done in parallel on a multi processor system, and parallel algorithms for the beamformation are evaluated. Test results show that the beamformation functions give accurate results. The run time for parallel beamformation of a two dimensional synthetic aperture image has been measured for 32, 16, 8, 4, 2 and one processors, respectively. The tests show that the run time is approximately halved, but not quite. The speedup is close to the desired value for two and four processors and quite close for eight processors. For more processors the speedup decreases more sharply, which is in agreement with Ahmdal’s law. Still, a more important reason for the decrease in speedup is probably that the network is more loaded when messages and data have to be sent between a larger number of clients. The speedup and efficiency is higher for larger input of data. The reason for this is probably that the relative time used for communication, compared to calculation, decreases. To say something about the scalability of the system more tests are needed, with different problem sizes and numbers of processors. Due to the high run times, the algorithm for three dimensional imaging has only been tested with 32, 16 and 8 processors, respectively. Therefore, it is not possible to say anything about speedup and efficiency in this case. The only thing that can be said is that the run time decreases when more processors are used.
Pages 101
Admin Creation date: 2009-06-08    Update date: 2009-06-08    Source: dtu    ID: 244308    Original MXD