Abstract

Ultrasound (US) imaging is increasingly attracting the attention of both academic and industrial researchers due to being a real-time and nonionizing imaging modality. It is also less
expensive and more portable compared to other medical imaging techniques. However, the
granular appearance hinders the interpretation of US images, hindering its wider adoption.
This granular appearance (also referred to as speckles) arises from the backscattered echo
from microstructural components smaller than the ultrasound wavelength, which are called
scatterers. While significant effort has been undertaken to reduce the appearance of speckles,
they contain scatterer properties that are highly correlated with the microstructure of the
tissue that can be employed to diagnose different types of disease. There are many properties
that can be extracted from speckles that are clinically valuable, such as the elasticity and
organization of scatterers. Analyzing the motion of scatterers in the presence of an internal
or external force can be used to obtain the elastic properties of the tissue. The technique
is called elastography and has been widely used to characterize the tissue. Estimating the
scatterer organization (scatterer number density and coherent to diffuse scattering power) is
also crucial as it provides information about tissue microstructure and potentially aids in disease diagnosis and treatment monitoring. This thesis proposes several deep learning-based
methods to facilitate and improve the estimation of speckle motion and scatterer properties, potentially simplifying the interpretation of US images. In particular, we propose new
methods for displacement estimation in Chapters 2 to 6 and introduce novel techniques in
Chapters 7 to 11 to quantify scatterers’ number density and organization.