Biomedical Imaging


Giovanni De Micheli
LSI Director and Professor


Federico Angiolini
Senior Scientific Collaborator
Aya Ibrahim
PhD Student

Ultrasound imaging

PhD project:

High Performance Portable 3D Ultrasound Platform


Contact person: Federico Angiolini, PhD, EPFL-IC-ISIM-LSI



Ultrasound imaging is an important biomedical technique for analyzing soft tissues in the human body, with both diagnostic and therapeutic applications. Ultrasound images are formed by emitting ultrasound waves from an array of piezoelectric transducers into the medium of interest, and then recording the echoes backscattered onto the same array. Beamforming techniques are then used to create an image from the received signals. Ultrasound imaging is the most widely-used medical imaging technique, because of its relative low cost, non-invasiveness and non-use of ionizing radiation, i.e. lack of adverse effects. It is widely used in prenatal care, for mammography and for many other applications (cardiac, renal, liver and gallbladder analysis, imaging of tendons, muscles and various superficial structures such as testicles, thyroid, etc.). Because of the real time nature of ultrasound, it is often used to guide surgical procedures as well.

i) Sonologist at work  ii) Conventional 2D ultrasound image  iii) Advanced 3D ultrasound imaging

Yet, ultrasound imaging has limitations. The quality of the resulting images is often poor if compared against more expensive procedures, such as Computed Tomography (CT) and Magnetic Resonance Imaging (MRI). Also, the image acquisition relies on manually operating the probe in direct contact with the patient’s skin, and experience and skill are required for the best diagnostic results – as opposed to automated scanning methods. For both reasons, trained sonologists must be in charge of operating the ultrasound scanners, rather than more generic medical personnel. Moreover, ultrasound imaging devices are usually targeted at hospital use, bulky and power-hungry, totally unsuitable for mobile and rescue applications, or for environments with unstable power supply. Miniaturized, lower-power ultrasound imaging devices have recently appeared on the market, but they provide medium quality at best


iv) Stationary ultrasound imaging apparatus  v) Industrial portable ultrasound device: GE Vscan

Project Objectives

UltrasoundToGo intends to develop a high-performance, low-power signal processing platform for ultrasound imaging applications, targeting future 3D portable ultrasound systems. The motivation of this work is to provide the means for achieving a portable medical system that can provide high-quality images while being battery operated, and thus much more usable in medical emergency or rescue operations.  It also paves the way for usage in areas, for example in some developing countries, with sporadic energy availability.

The improved image quality, the volumetric 3D scanning, and the flexibility of the platform are intended to make ultrasound imaging devices much easier to use also by non-specifically-trained personnel. This enables telemedicine scenarios, where high-quality scans could be effortlessly and safely acquired by general practitioners, and then uploaded to remote specialists for further analysis if necessary.


High-quality volumetric rendering of a heart valve


Technology and Preliminary Results

UltrasoundToGo relies on innovation from the hardware, algorithmic and software implementation side.

From the hardware side, UltrasoundToGo focuses on providing extremely high-bandwidth signal processing and advanced computing capabilities in a low-power envelope, compatible with battery operation. The problem’s core is being tackled as an FPGA design at the LSI lab, while the design of a scalable ASIC with breakthrough performance/power has been completed at ETHZ’s IIS lab. A key challenge is to intelligently manage data bandwidth requirements, which in a naive implementation would reach multiple TB/s.


Left: On-the-fly calculation of coefficient tables  Right: Architecture of beamforming ASIC chip in 28nm


From the algorithmic side, the LTS5 lab has been approaching the image reconstruction problem from a novel angle, deploying compressive sensing techniques to the medical ultrasound domain. The outcome is an increase in image contrast and a reduction in the number of insonifications required to reconstruct each frame.

Left: Higher contrast is possible with fewer insonification (pink line)  Right: Contrast improvement in a real carotid scan

From the software side, UltrasoundToGo’s RISD  and TIK  labs are creating innovative highly parallel programming approaches that suit the nature of ultrasound image processing. Distinctive features will include a qualified software deployment and maintenance model whereby new real-time control and analysis algorithms can be downloaded on the platform infield, under end-user control. This model is supported by a formally well-defined and sound programming model and verification methodology for guaranteeing correctness and quality of results.

Left: Novel application mapping approach  Right: Memory savings from customized memory management

Project Partners
UltrasoundToGo is a funded project. UltrasoundToGo involves the LSI, LTS5, and RISD Laboratories at the École Polytechnique Fédérale de Lausanne (EPFL), and IIS and TIK Laboratories at the Eidgenössische Technische Hochschule Zürich (ETHZ), and collaborates with the Centre Hospitalier Universitaire Vaudois (CHUV) and international academic and industrial partners.



Pirmin Vogel, Andrea Bartolini, Luca Benini, “Efficient Parallel Beamforming for 3D Ultrasound Imaging”, GLSVLSI’14 Proceedings of the 24th edition of the great lakes symposium on VLSI, Houston, Texas, USA, pp. 175-180, May 2014.

Stefanos Skalistis, Alena Simalatsar, “Modeling of Reconfigurable Medical Ultrasonic Applications in BIP”, Proceedings of the 5th Workshop on Medical Cyber-Physical Systems, Dagstuhl, Germany, pp. 66-79, 2014.

Aya Ibrahim, Pascal Hager, Andrea Bartolini, Federico Angiolini, Marcel Arditi, Luca Benini, Giovanni De Micheli, “Tackling the Bottleneck of Delay Tables in 3D Ultrasound Imaging”, Proceedings of the DATE Conference, Grenoble, France, 2015, pp. 1683-1688.

Aya Ibrahim, Alena Simalatsar, Stefanos Skalistis, Federico Angiolini, Marcel Arditi, Jean-Philippe Thiran, Giovanni De Micheli, “Assessment of Image Quality vs. Computation Cost for Different Parameterizations of Ultrasound Imaging Pipelines”, Proceedings of the 6th Workshop on Medical Cyber-Physical Systems, Seattle, USA, April 13th 2015.

Andreas Tretter, Harshavardhan Pandit, Pratyush Kumar, Lothar Thiele, “Deterministic Memory Sharing in Kahn Process Networks: Ultrasound Imaging as a Case Study”, Proceedings of the 2014 IEEE 12th Symposium on Embedded Systems for Real-time Multimedia (ESTIMedia), pp. 80-89, 2014.

Pascal Alexander Hager, Pirmin Vogel, Andrea Bartolini, Luca Benini, “Assessing the Area/Power/Performance Tradeoffs for an Integrated Fully-Digital, Large-Scale 3D-Ultrasound Beamformer”, Proceedings of the 2014 IEEE Biomedical Circuits and Systems Conference (BioCAS), pp. 228-231, Lausanne, CH, Oct 2014.

O. Bernard, M. Zhang, F. Varray, J. P. Thiran, H. Liegbott, D. Friboulet, “Ultrasound Fourier Slice Imaging: a novel approach for ultrafast imaging technique”, Proceedings of the 2014 IEEE International Ultrasonics Symposium (IUS), pp. 129-132, Oct 2014.

Andreas Tretter, Pratyush Kumar, Lothar Thiele, “Interleaved Multi-Bank Scratchpad Memories: A Probabilistic Description of Access Conflicts”, Proceedings of the 52nd DAC Conference, San Francisco, CA, USA, June 2015, p. 22.