Segmentation of ultrasound image data by two dimensional autoregressive modelling

Abstract

In this paper we treat ultrasound image data as a two dimensional autoregressive (AR) signal. The image is modelled as consisting of distinct regions each described by one of a small number of AR models. Segmentation is performed by maximising the image likelihood function, which takes on a convenient form due to the AR model. Image data is presented to the algorithm in complex amplitude form. Results from application of this method to a cardiac phantom data set are presented.