This thesis presents acceleration techniques for medical imaging algorithms. The rapid development of medical scanning devices produces huge amounts of raw data. On the one hand, high-resolution images can be computed from the raw data and, thus,
providing the physicians better basis for diagnosis. On the other hand, the amount of
raw data leads to longer processing times. About three years ago, graphics processing
units (GPUs) have become programmable and can be used for other tasks than graphics.
GPUs are very suitable for medical imaging algorithms – they are fast, relatively cheap,
the instruction set is powerful enough for many of algorithms, and with standard APIs
easy to program. In this work, medical imaging filtering, reconstruction, segmentation,
and registration algorithms are accelerated using GPUs. Averagely a 10-times speedup
is achieved compared to optimized SSE3 CPU implementations. Some implementations
run 100 times faster than their CPU counterparts. The results are already used
successfully in products by Siemens Medical Solutions.
«
This thesis presents acceleration techniques for medical imaging algorithms. The rapid development of medical scanning devices produces huge amounts of raw data. On the one hand, high-resolution images can be computed from the raw data and, thus,
providing the physicians better basis for diagnosis. On the other hand, the amount of
raw data leads to longer processing times. About three years ago, graphics processing
units (GPUs) have become programmable and can be used for other tasks than gra...
»