LightProbe - Implementation of compressed-sensing algorithms
Ultrasound imaging is one of the most important diagnostic procedures in modern medicine. Being non-invasive and not emitting any potentially harmful radiation (like X-Rays), it is relatively low-cost at the same time. Yet, it could be used in many more contexts and situations still. This is however hampered by the size and the power-consumption of current such systems. While there are also mobile ultrasound devices, their image quality is distinctively worse.
The swiss-founded Nano-Tera "Ultrasound To Go" project now tries to develop portable high-quality and low-power ultrasound devices, which would allow for ultrasound diagnostics in a new field of situations, e.g. in emergency cases or in rural areas. A new and promising approach, which we are investigating, is to use compressed sensing techniques, which at the price of higher computational complexity allow to achieve high image quality with few inputs (and thus a higher time resolution).
Compressed sensing (CS) introduces a signal acquisition framework that goes beyond the traditional Nyquist sampling paradigm. Under strict conditions on the measurement process and structural assumptions on signals under scrutiny, CS demonstrates that signals can be acquired using a small number of linear measurements and then recovered by solving a non-linear optimization problem. Thus, the efficiency of CS is intrinsiquely conditioned by the ability to design non-standard acquisition schemes and the existence of algorithms able to solve the problem posed by CS.
LTS 5 at EPFL (http://lts5www.epfl.ch/) has developed different algorithms based on the CS framework which are able to recover high quality images with fewer measurements than classical methods. In parallel, IIS at ETHZ (http://www.iis.ee.ethz.ch/) is currently developing the next generation of ultrasound probe (used to acquire ultrasound signals), called LightProbe, in which the signal is digitized directly in the probe and sent to the US system through an optical fiber. This digitization enables a higher flexibility for the design of acquisition strategies than classical probes which operate on analog signals.
During the thesis, which is offered collaboratively between IIS and LTS 5, the student will bridge the gap between the algorithms developed at LTS 5 and the probe designed at IIS by studying CS strategies that may be implemented in the LightProbe. This involves the following steps:
- Understand the CS framework: theoretical principles and hardware implementation
- Familiarize himself with ultrasound imaging: signal acquisition and image reconstruction
- Familiarize himself with the LightProbe
- Choose a CS strategy to be implemented in the LightProbe
- Hardware Evaluations
- Implement and test the strategy
- Looking for Interested Students
- Supervision: Pascal Hager, Adrien Besson (firstname.lastname@example.org)
- 30% Understanding Theory
- 40% Evaluations
- 30% Implementation
- Knowledge of signal processing and digital hardware design (VLSI I).
- Skills in MATLAB.
- Prof. Jean-Philippe Thiran (LTS 5 - EPFL)
- Prof. Luca Benini (IIS - ETH)