|M.Sc Student||Yael Yankelevsky|
|Subject||Sparse Representation of Ultrasound Signals using Raw-Data|
|Department||Department of Electrical Engineering||Supervisors||Professor Emeritus Feuer Arie|
|Dr. Zvi Friedman|
|Full Thesis text|
Medical ultrasound imaging is frequently used due to its many advantages: it is safe to the patient, non-invasive in its nature, compact and portable, relatively affordable, allows real time visualization of moving objects, and suitable for many clinical applications. However, ultrasound images typically suffer from poor quality compared to other medical imaging modalities such as Magnetic Resonance Imaging (MRI) and Computed Tomography (CT), partly due to side-lobes artifacts that degrade the contrast of the resulting image and tend to obscure and mask diagnostically important details. Therefore, improving the quality of the ultrasound image is a great challenge.
This challenge is contradicted by another challenge of reducing the enormous amounts of data involved.
Modern ultrasound systems require
large amounts of data to be acquired, transferred and processed. Advancements
in technology accompanied by growing imaging demands require that these large
amounts of data be employed in more computational tasks. Therefore, it would be
desirable to reduce the amount of needed data without compromising the image
quality and its diagnostic credibility.
The combined goal of reducing the amount of data without reducing, or even improving, the image quality, is a real challenge of much importance.
Motivated by the above, we propose a representation and processing scheme for ultrasonic signals based on the decomposition of the received signals into two components that can each be sparsely represented. An independent compression of each component individually is shown to yield better overall data reduction ratios compared with direct compression of the detected signal.
Furthermore, applying this decomposition at early stages of the imaging process, namely before the receive beamforming step, we show that the aforementioned side-lobes artifacts are significantly improved alongside the data size reduction, thus improving the image quality.
The effectiveness of the proposed solution is demonstrated experimentally on simulated data as well as on real cardiac ultrasound data.
Applying our processing scheme to cardiac ultrasound data, we successfully detect the macroscopic perturbations in the tissue and also maintain the tissue's texture information, while achieving an over twenty-fold reduction in data size.