Ultrasound characterization by stable statistical patterns.

It is shown that automated characterization of ultrasound images can be achieved effectively if the images are confined to single internal organs. The ultrasound image of a particular organ gives statistical textures of its structures, which are consistent with stable patterns. Analysis of these patterns leads to the construction, from a few statistical parameters, of a fast and effective algorithm to characterize these ultrasound images. A small training set is adequate. A similar approach may be used to detect tumour sites from liver ultrasonograms. The algorithms are implemented on an IBM-compatible personal computer for future incorporation into existing ultrasound machines.