High Frequency Ultrasound RF Time Series Analysis for Tissue Characterization

Abstract

Ultrasound-based tissue characterization has been an active field of cancer detection in the past decades. The main concept behind various techniques is that the returning ultrasound echoes carry tissue-dependent information that can be used to distinguish tissue types. Recently, a new paradigm for tissue typing has been proposed which uses ultrasound Radio Frequency (RF) echoes, recorded continuously from a fixed location of the tissue, to extract tissue-dependent information. This is hereafter referred to as RF time series. The source of tissue typing information in RF time series is not a well known concept in the literature. However, there are two main hypotheses that describe the informativeness of variations in RF time series. Such information could be partly due to heat induction as a result of consistent eradiation of tissue with ultrasound beams which results in a virtual displacement in RF echoes, and partly due to the acoustic radiation force of ultrasound beams resulting in micro-vibration inside tissue. In this thesis, we further investigate RF time series signals, collected at high frequencies, by analyzing the properties of the RF displacements. It will be shown that such displacements exhibit oscillatory behavior, emphasizing on the possible micro-vibrations inside tissue, as well as linear incremental trend, indicating the effect of heat absorbtion of tissue. The main focus of this thesis is to study the oscillatory behavior of RF displacements in order to extract tissue-dependent features based on which tissue classification is performed. Using various linear and nonlinear tools, we study the properties of such displacements in both frequency and time domain. Nonlinear analysis, based on the theory of dynamical systems, is used to study the dynamical and geometrical properties of RF displacements in the time domain. Using Support Vector Machine (SVM), different tissue typing experiments are performed to investigate the capability of the proposed features in tissue classification. It will be shown that the combination of such features can distinguish between different tissue types almost perfectly. In addition, a feature reduction algorithm, based on principle component analysis (PCA), is performed to reduce the number of features required for a successful tissue classification.