Abstract-Lossy image compression algorithms are pervasively used to reduce the size of images transmitted over the web and recorded on data storage media. However, we pay for their high compression rate with visual artifacts degrading the user experience. Deep convolutional neural networks have become a widespread tool to address high-level computer vision tasks very successfully. Recently, they have found their way into the areas of low-level computer vision and image processing to solve regression problems mostly with relatively shallow networks.We present a novel 12-layer deep convolutional network for image compression artifact suppression with hierarchical skip connections and a multi-scale loss function. We achieve a boost of up to 1.79 dB in PSNR over ordinary JPEG and an improvement of up to 0.36 dB over the best previous ConvNet result. We show that a network trained for a specific quality factor (QF) is resilient to the QF used to compress the input image-a single network trained for QF 60 provides a PSNR gain of more than 1.5 dB over the wide QF range from 40 to 76.

Abstract-3D ultrasound imaging is quickly becoming a reference technique for high-quality, accurate, expressive diagnostic medical imaging. Unfortunately, its computation requirements are huge and, today, demand expensive, power-hungry, bulky processing resources. A key bottleneck is the receive beamforming operation, which requires the application of many permutations of fine-grained delays among the digitized received echoes. To apply these delays in the digital domain, in principle large tables (billions of coefficients) are needed, and the access bandwidth to these tables can reach multiple TB/s, meaning that their storage both on-chip and off-chip is impractical. However, smarter implementations of the delay generation function, including forgoing the tables altogether, are possible. In this paper we explore efficient strategies to compute the delay function that controls the reconstruction of the image, and present a feasibility analysis for an FPGA platform.

Abstract-Ultrasound imaging is a reference medical diagnostic technique thanks to its blend of versatility, effectiveness and moderate cost. The core computation of all ultrasound imaging methods is based on simple formulae, except for those required to calculate delays with high precision and throughput. Unfortunately, advanced 3D systems require the calculation or storage of billions of such delay values per frame, which is a challenge. In 2D systems, this requirement can be four orders of magnitude lower, but efficient computation is still crucial in view of low-power implementations that can be battery-operated, enabling usage in rescue scenarios.In this paper we explore two smart designs of the delay generation function. To quantify their hardware cost, we implement them on FPGA and study their footprint and performance. We evaluate how these architectures scale to different ultrasound applications, from a low-power 2D system to a next-generation 3D machine. When using numerical approximations, we demonstrate the ability to generate delay values with sufficient throughput to support 10000-channel 3D imaging at up to 30 fps while using 63% of a Virtex 7 FPGA, requiring 24 MB of external memory accessed at about 32 GB/s bandwidth. Alternatively, with similar FPGA occupation, we show an exact calculation method that reaches 24 fps on 1225-channel 3D imaging and does not require external memory at all. Both designs can be scaled to use a negligible amount of resources for 2D imaging in low-power applications, and for ultrafast 2D imaging at hundreds of frames per second.

High-frame-rate and high-resolution 3D medical ultrasound imaging imposes high requirements on the involved processing hardware. Several thousands of analog signals need to be processed in many steps to obtain a final image. Fully digital beamforming makes it possible to achieve high image quality coupled with extreme flexibility. Unfortunately, digital beamforming imposes staggering requirements on main memory bandwidth caused by the loading of off-chip stored beamforming delays. In this paper we present the first fully-digital integrated beamformer that is able to compute 269.3 M focal points (FP) per second from 10 000 receive channels, and which does not require off-chip main memory. This is enabled by our novel delay approximation circuit that exploits temporal correlation between subsequent computations and thereby allows to compute the delays for beamforming online. To estimate the area and power requirements, the complete system was designed and the beamformer core was evaluated for a 130 nm CMOS technology. The estimated complexity per channel is 37.2 kGE and the corresponding power dissipation was estimated with 48 mW.

Digital ultrasound probes integrate the analog frontend in the housing of the probe handle and provide a digital interface instead of requiring an expensive coaxial cable harness to connect. Current digital probes target the portable market and perform the bulk of the processing (beamforming) on the probe, which enables the probe to be connected to commodity devices such as tablets or smartphones running an ultrasound app to display the image and control the probe. Thermal constraints limit the number of front-end channels as well as the complexity of the processing. This prevents current digital probes to support advanced modalities such as Vector Flow or Elastography requiring high-frame rate (HFR) imaging. In this paper, we present LIGHTPROBE, a digital ultrasound probe, which integrates a 64-channel 100 Vpp TX/RX frontend including analog-to-digital conversion (up to 32.5 MS/s @ 12 bit), and is equipped with an optical high-speed link (26.4 Gb/s) providing sustainable raw samples access to all channels, which allows the processing to be performed on the connected device without thermal power constraints. By connecting the probe to a GPU-equipped PC, we demonstrate the flexibility of softwaredefined B-mode imaging using conventional and ultrafast methods. We achieve plane-wave and synthetic aperture imaging with frame-rates from 30 Hz up to 500 Hz consuming between 5.6 W and 10.7 W. By using a combination of power and thermal management techniques, we demonstrate that the probe can remain within operating temperature limits even without active cooling, while never having to turn the probe off for cooling hence providing a consistent Quality of Service for the operator.