|M.Sc Student||Cohen Yossef|
|Subject||On Model-Based Analysis and Representation of Ultrasound|
|Department||Department of Electrical Engineering||Supervisors||Professor Moshe Porat|
|Dr. Zvi Friedman|
|Full Thesis text|
Ultrasound is a widely used medical imaging tool. An ultrasound image is commonly formed using Delay-and-Sum (DAS) beamforming with uniform weighting. The main disadvantages of DAS beamforming are wide main-lobe and significant side-lobes, which lead to low resolution and contrast of the image. In particular, strong reflectors that are received through the beam’s side-lobe override and mask weak reflectors that are received through the main-lobe. To deal with the drawbacks of the DAS method, previous works suggested using adaptive beamforming methods. Adaptive beamforming uses a weighted sum of the sampled data, whereas the different weights are selected for each imaged point according to a quality criterion. These methods succeed in improving the angular resolution, yet the destructive influence of the strong reflections is not fully removed. This research proposes a novel method, which is applied prior to the beamforming step and allows to significantly reduce the effect of the strong reflections received through the side-lobes. Using a sparsity-prior for the strong reflectors, we propose a method for locating strong reflectors and separating them from the US image. The method is based on the CLEAN algorithm. CLEAN is an iterative algorithm, originally implemented on radio astronomy signals, which deconvolves a sampling function (the “Dirty Beam”) from an observed transmission (the “Dirty Map”) of a radio source. To the best of our knowledge, this is the first successful attempt to implement the CLEAN algorithm on ultrasonic images.
An image based on the weak reflectors is then built using Low Complexity Adaptive (LCA) beamforming. Lastly, the strong reflectors are artificially added back to the ultrasound image. Based on 10,000 in-silico simulations we show that the location and reflectivity of a sole randomly located strong reflector are extracted with good accuracy for reflectivity values which are 13dB up to 33dB higher than its environment. We further show using in-silico and in-vivo examples that after subtracting the contribution of the strong
reflectors, the reflectivity function of the weaker reflectors is faithfully recovered via adaptive beamforming. Our results show that the proposed model-based approach to ultrasound imaging can significantly enhance ultrasound imaging quality.