Matthias Birk

A comprehensive comparison of GPU- and FPGA-based acceleration of reflection image reconstruction for 3D ultrasound computer tomography

As today’s standard screening methods frequently fail to diagnose breast cancer before metastases have developed, earlier breast cancer diagnosis is still a major challenge. Three-dimensional ultrasound computer tomography promises high-quality images of the breast, but is currently limited by a time-consuming image reconstruction. In this work, we investigate the acceleration of the image reconstruction by GPUs and FPGAs. We compare the obtained performance results with a recent multi-core CPU. We show that both architectures are able to accelerate processing, whereas the GPU reaches the highest performance. Furthermore, we draw conclusions in terms of applicability of the accelerated reconstructions in future clinical application and highlight general principles for speed-up on GPUs and FPGAs.

3D ultrasound computer tomography of the breast: A new era?

A promising candidate for imaging of breast cancer is ultrasound computer tomography (USCT). The main advantages of a USCT system are simultaneous recording of reproducible reflection, attenuation and speed of sound volumes, high image quality, and fast data acquisition. The here presented 3D USCT prototype realizes for the first time the full potential of such a device. It is ready for a clinical study. Full volumes of a breast can be acquired in four minutes. In this paper images acquired with a clinical breast phantom are presented. The resolution and imaged details of the reflectivity reconstruction are comparable to a 3 tesla MRI volume of the phantom. Image quality and resolution is isotropic in all three dimensions, confirming the successful implementation experimentally.

Hardware setup for the next generation of 3D Ultrasound Computer Tomography

We describe the second generation of a 3D-Ultrasound Computer Tomography (USCT) system. After we achieved in the first generation a device with sub-wavelength resolution and three imaging modalities (reflection, attenuation, speed of sound) and tested it with static phantoms, we developed a device for in-vivo imaging. In the new system the geometry of transducers and their spatial distribution is optimized in respect to uniformity and high value of: contrast, resolution, and illumination. Furthermore we developed new electronics which allows faster DAQ (≤ 2 min) and contains larger and faster FPGAs to use their processing power for data pre-processing.

GPU-based iterative transmission reconstruction in 3D ultrasound computer tomography

As todays standard screening methods frequently fail to detect breast cancer before metastases have developed, early diagnosis is still a major challenge. With the promise of high-quality volume images, three-dimensional ultrasound computer tomography is likely to improve this situation, but has high computational needs. In this work, we investigate the acceleration of the ray-based transmission reconstruction by a GPU-based implementation of the iterative numerical optimization algorithm TVAL3. We identified the regular and transposed sparse-matrix-vector multiply as the performance limiting operations. For accelerated reconstruction we propose two different concepts and devise a hybrid scheme as optimal configuration. In addition we investigate multi-GPU scalability and derive the optimal number of devices for our two primary use-cases: a fast preview mode and a high-resolution mode. In order to achieve a fair estimation of the speedup, we compare our implementation to an optimized CPU version of the algorithm. Using our accelerated implementation we reconstructed a preview 3D volume with 24,576 unknowns, a voxel size of (8?mm)3 and approximately 200,000 equations in 0.5?s. A high-resolution volume with 1,572,864 unknowns, a voxel size of (2mm)3 and approximately 1.6 million equations was reconstructed in 23?s. This constitutes an acceleration of over one order of magnitude in comparison to the optimized CPU version. We accelerate a ray-based 3D ultrasound CT reconstruction by GPU processing.By use-case optimized SpMV variants a speedup of one order of magnitude is obtained.We derive the optimal number of GPUs for reconstructions that do not fit on one GPU.A 3D-preview is reconstructed in 0.5 s, a high-resolution volume in 23 s.

GPU based acceleration of 3D USCT image reconstruction with efficient integration into MATLAB

3D ultrasound computer tomography (3D USCT) promises reproducible high-resolution images for early detection of breast tumors. The synthetic aperture focusing technique (SAFT) used for image reconstruction is highly computeintensive but suitable for an accelerated execution on GPUs. In this paper we investigate how a previous implementation of the SAFT algorithm in CUDA C can be further accelerated and integrated into the existing MATLAB signal and image processing chain for 3D USCT. The focus is on an efficient preprocessing and preparation of data blocks in MATLAB as well as an improved utilisation of special hardware like the texture fetching units on GPUs. For 64 slices with 1024×1024 pixels each the overall runtime of the reconstruction including data loading and preprocessing could be decreased from 35 hours with CPU to 2.4 hours with eight GPUs.

Reconfigurable MPSoC versus GPU: Performance, power and energy evaluation

Different characteristics of algorithms, perform better or worse on various target hardware. The consequent of this is, that the selection of one suitable hardware, such as Graphic Processing Units (GPU), Field Programmable Gate Arrays (FPGAs) or traditional processor cores is a challenging task for developers. The challenge is to choose the most suitable platform satisfying the requirements of the given application, such as real-time and power- / energy consumption constraints. Due to short time to market pressure, a fast development cycle is also a very important characteristic in industrial applications. This work evaluates the performance, power- /energy consumption and the development effort for three processor-based systems: an FPGA-based multiprocessor platform called RAMPSoC, a GPU and a CPU. Two real-world applications have been selected for this evaluation: 3D Ultrasound Computer Tomography and Object Recognition. The paper presents the realization of the applications on the different target hardware and discusses the results of the power and performance evaluation.

Acceleration of image reconstruction in 3D ultrasound computer tomography: An evaluation of CPU, GPU and FPGA computing

As todays standard screening methods frequently fail to diagnose breast cancer before metastases have developed, earlier breast cancer diagnosis is still a major challenge. Three-dimensional ultrasound computer tomography promises high-quality images of the breast, but is currently limited by a time-consuming synthetic aperture focusing technique based image reconstruction. In this work, we investigate the acceleration of the image reconstruction by a GPU, and by the FPGAs embedded in our custom data acquisition system. We compare the obtained performance results with a recent multi-core CPU and show that both platforms are able to accelerate processing. The GPU reaches the highest performance. Furthermore, we draw conclusions in terms of applicability of the accelerated reconstructions in future clinical application and highlight general principles for speed-up on GPUs and FPGAs.

Evaluation of performance and architectural efficiency of FPGAs and GPUs in the 40 and 28nm generations for algorithms in 3D ultrasound computer tomography

In heterogeneous computing, application developers have to identify the best-suited target platform from a variety of alternatives. In this work, we compare performance and architectural efficiency of Graphics Processing Units (GPUs) and Field Programmable Gate Arrays (FPGAs) for two algorithms taken from a novel medical imaging method named 3D ultrasound computer tomography. From the 40nm and 28nm generations, we use top-notch devices and those with similar power consumption values. For our two benchmark algorithms from the signal processing and imaging domain, the results show that if power consumption is not considered, the GPU and FPGA from the 40nm generation give both, a similar performance and efficiency per transistor. In the 28nm process, in contrast, the FPGA is superior to its GPU counterpart by 86% and 39%, depending on the algorithm. If power is limited, FPGAs outperform GPUs in each investigated case by at least a factor of four.

Comparison of processing performance and architectural efficiency metrics for FPGAs and GPUs in 3D Ultrasound Computer Tomography

With the rise of heterogeneous computing architectures, application developers are confronted with a multitude of hardware platforms and the challenge of identifying the most suitable processing platform for their application. Strong competitors for the acceleration of 3D Ultrasound Computer Tomography, a medical imaging method for early breast cancer diagnosis, are GPU and FPGA devices. In this work, we evaluate processing performance and efficiency metrics for current FPGA and GPU devices. We compare top-notch devices from the 40 nm generation as well as FPGA and GPU devices, which draw the same amount of power. For our two benchmark algorithms, the results show that if power consumption is not considered the GPU and the FPGA give both, a similar processing performance and processing efficiency per transistor. However, if the power budget is limited to a similar value, the FPGA performs between six and eight times better than the GPU.

High-Speed Medical Imaging in 3D Ultrasound Computer Tomography

A promising candidate for sensitive imaging of breast cancer is 3D Ultrasound Computer Tomography (3D USCT). So far its clinical applicability for diagnosis has been limited by the duration of the demanding image reconstruction. In this paper we investigate how signal processing and image reconstruction can be accelerated for diagnosis by using heterogeneous hardware. Additionally, the time and costs for real-time system for a future diagnosis and therapy device is estimated. Reusing the devices built-in FPGA-based data acquisition system (DAQ) through reconfiguration results in a speed-up by a factor of 7 for signal processing and by a factor of 2 for image reconstruction. Applying cutting-edge single FPGAs and GPUs, speed-ups by a factor of 10 (FPGA) and 6 (GPU) for signal processing and 15 (FPGA) and 37 (GPU) for image reconstruction were achieved compared to a recent quad-core Intel Core-i7 CPU. Using quad-core CPUs and a cluster of eight GPUs allowed us for the first time to calculate volumes in less than 30 min with an overall speed-up by a factor of 47, enabling a first clinical study. Based on these results we extrapolated that real-time reconstruction for a therapeutic 3D USCT will be possible in the year 2020 if the trend in density follows the ITRS roadmap.