Breast tissue characterization based on fractional differencing model of ultrasonic RF echo

Abstract

A number of researchers have previously shown that the ultrasound RF echo of tissue exhibits (1/ f ) -β characteristics and developed tissue characterization methods based on the fractal parameter β. In this paper we propose Fractional Differencing Autoregressive Moving Average (FARMA) process for modeling RF ultrasound echo and develop breast tissue characterization method based on the FARMA model parameters. This model has been used to capture statistical self-similarity and long-range correlations in image textures, in wide ranging engineering and science applications, including communication network traffic. Here, we present estimation techniques to extract the model parameters, namely features, for classification purposes and tissue characterization. We show the performance of our tissue characterization procedure on several in vivo ultrasound breast images including benign and malignant tumors. The area of the receiver operator characteristics (ROC) based on 60 in vivo images yields a value of 0.79, which indicates that proposed tissue characterization method is comparable in performance with other successful methods reported in the literature.