Browsing by Author "Schaetz, Sebastian"

We present MGPU, a C++ programming library targeted at single-node multi-GPU systems. Such systems combine disproportionate floating point performance with high data locality and are thus well suited to implement real-time algorithms. We describe the library design, programming interface and implementation details in light of this specific problem domain. The core concepts of this work are a novel kind of container abstraction and MPI-like communication methods for intra-system communication. We further demonstrate how MGPU is used as a framework for porting existing GPU libraries to multi-device architectures. Putting our library to the test, we accelerate an iterative non-linear image reconstruction algorithm for real-time magnetic resonance imaging using multiple GPUs. We achieve a speed-up of about 1.7 using 2 GPUs and reach a final speed-up of 2.1 with 4 GPUs. These promising results lead us to conclude that multi-GPU systems are a viable solution for real-time MRI reconstruction as well as signal-processing applications in general.

To develop generic optimization strategies for image reconstruction using graphical processing units (GPUs) in magnetic resonance imaging (MRI) and to exemplarily report on our experience with a highly accelerated implementation of the nonlinear inversion (NLINV) algorithm for dynamic MRI with high frame rates.

PURPOSE: To evaluate a novel real-time phase-contrast magnetic resonance imaging (MRI) technique for the assessment of through-plane flow in the ascending aorta. MATERIALS AND METHODS: Real-time MRI was based on a radial fast low-angle shot (FLASH) sequence with about 30-fold undersampling and image reconstruction by regularized nonlinear inversion. Phase-contrast maps were obtained from two (interleaved or sequential) acquisitions with and without a bipolar velocity-encoding gradient. Blood flow in the ascending aorta was studied in 10 healthy volunteers at 3 T by both real-time MRI (15 sec during free breathing) and electrocardiogram (ECG)-synchronized cine MRI (with and without breath holding). Flow velocities and stroke volumes were evaluated using standard postprocessing software. RESULTS: The total acquisition time for a pair of phase-contrast images was 40.0 msec (TR/TE = 2.86/1.93 msec, 10° flip angle, 7 spokes per image) for a nominal in-plane resolution of 1.3 mm and a section thickness of 6 mm. Quantitative evaluations of spatially averaged flow velocities and stroke volumes were comparable for real-time and cine methods when real-time MRI data were averaged across heartbeats. For individual heartbeats real-time phase-contrast MRI resulted in higher peak velocities for values above 120 cm s(-1). CONCLUSION: Real-time phase-contrast MRI of blood flow in the human aorta yields functional parameters for individual heartbeats. When averaged across heartbeats real-time flow velocities and stroke volumes are comparable to values obtained by conventional cine MRI.