PrintParMRI - Fully Automated Tumor Diagnosis for MRI
The Project
Magnet Resonance Imaging (MRI) is an important part of current internal medicine, allowing non-invasive examinations to diagnose and treat medical conditions. One major field of application is the breast cancer and tumour detection. A tool-flow of different image processing algorithms is used to support the radiologists identifying and rating tumour and cancer cells. These tools include image registration, image segmentation and different methods of feature extraction and classification. All of these algorithms work on large amounts of data and experience a long runtime based on complex algorithms. In this project we aim on developing new algorithms and finding customized solutions to accelerate the tool-flow of tumour diagnosis for breast-MRI.
Our Mission
The main approach followed is to parallelize the used and newly developed algorithms using current and new parallelization techniques on diverse parallel hardware. The hardware used for this purpose is aimed to be suited best possible to the set of algorithms. Thus, a combination of different parallel architectures is incorporated for speeding up the complete tool-flow. These architectures especially include current Many-Core architectures and Graphic Processor Units (GPUs), but also more specialized and less common hardware like Field Programmable Gate Arrays (FPGAs) and other co-processors like ClearSpeed accelerators.
The role of the PC² is to consult the CADMEI GmbH during the algorithm development by the design of data structures, which map very well on parallel architectures. Moreover, we will accelerate the developed algorithms on appropriate architectures and design a hardware product setup, which will be optimized in terms of cost-benefit ratio.
Partners
Funding
- BMWi - Zentrales Innovationsprogramm Mittelstand (ZIM)
Duration
- 07/2009 – 06/2011
