Estimation algorithms for quantitative tissue characterization in ultrasound images using doubly stochastic translated point processes

Abstract

Quantitative biological tissue characterization is of paramount importance for computer-assisted tissue classification and medical diagnosis. This paper presents stochastic models and estimation algorithms for the average local density or concentration of scatterers in tissues using speckled ultrasound images. We treat speckle form a novel point of view: as a carrier of useful clinical information about tissue characteristics rather than as contaminating noise. The stochastic models for tissue scattering are based on a doubly stochastic compound marked Poisson point process. For each of these tissue scattering statistical models, we present estimation algorithms to determine the average tissue local scatterer density, using intensity measurements of speckled images. We show that the maximum likelihood estimator is optimal in the sense that the variance of the error is the smallest possible using any other conceivable estimate having the same bias with the same data. In addition to their important applications in biological tissue classification, these estimation algorithms serve as a powerful tool for estimating radioactive concentration and for image reconstruction in tomography.