Dipl.-Inf. Hannes Hofmann
Alumnus of the Pattern Recognition Lab of the Friedrich-Alexander University Erlangen-Nuremberg
He who stops being faster stops being fast.
Field of Research: Hardware accelerated medical image processing
Most medical image processing tasks are computationally very demanding and involve huge data sets. However, in a clinical environment, runtime sometimes really matters. Therefore, the latest high performance hardware architectures are used to accelerate processing of the data.
The focus of my PhD thesis is accelleration of medical image reconstruction algorithms using Intel's Larrabee many-core architecture. For further information about Larrabee I shall refer you to an extensive article on Anandtech and its Wikipedia article. Together with our CUDA Guy Benni I'm working on Hardware Acceleration Techniques for 3D Cone-Beam Reconstruction.
The goal of this research is to enable and optimize the imaging toolchain on upcoming hardware platforms. The algorithms I investigate on are reconstructions for transmission (X-ray CT) as well as magnetic resonance imaging (MRI) and emission tomography (SPECT) .
If you are interested in working with Larrabee or another Multi-core architecture in your thesis, please contact me. There's always a topic available for motivated students.
There are many contributions to high performance CT reconstruction. However, the published algorithms' runtimes are often not comparable due to proprietary acquisition setups and different data sets. Therefore, the Pattern Recognition Lab initiated RabbitCT, a benchmark for C-arm CT reconstruction algorithms. We provide a standardized preprocessed dataset that can be used to evaluate the performance of reconstruction implementations. Recently, Christopher, Benni and I launched the RabbitCT website and started the race for the fastest backprojection. An article describing all details was published in Medical Physics (see pubs page).
During my study thesis we worked on the development of a framework to leverage image reconstruction (and arbitrary image processing) tasks on a variety of hardware platforms. Besides the framework development my task was to optimize two image filters on IBM's CELL processor (CBEA).