B-scan texture classification: a study using physical and theoretical models.

Abstract

Recent clinical trials, in our own centre and elsewhere, have demonstrated the ability of A- and B-scan texture analysis techniques to provide useful in vivo separation of certain disease conditions. The present paper reports studies, using physical and theoretical models of tissues, aimed at identifying the particular sets of textural features which are most effective in achieving such separations. Using a 3.5 MHz, single element manual scanner, a total of 300 B-scans have been collected from four physical models consisting of gel-embedded 'point' scatterers, with differing mean separations. Additionally, computer modelling has been used to obtain a further 500 B-scans from 14 statistically different scattering situations. The total set of B-scans thus obtained has been subjected to a variety of texture classification procedures which have previously been developed for application in other imaging fields, such as cytology and aerial photography. The results are discussed in terms of: i) the features which are best suited to the classification of B-scans, and ii) the manner in which the resulting classification schemes are sensitive to variables of the ultrasonic scanning instrumentation.