Errors in biased estimators for parametric ultrasonic imaging.

Abstract

Maximum likelihood (ML) methods are widely used in acoustic parameter estimation. Although ML methods are often unbiased, the variance is unacceptably large for many applications, including medical imaging. For such cases, Bayesian estimators can reduce variance and preserve contrast at the cost of an increased bias. Consequently, including prior knowledge about object and noise properties in the estimator can improve low-contrast target detectability of parametric ultrasound images by improving the precision of the estimates. In this paper, errors introduced by biased estimators are analyzed and approximate closed-form expressions are developed. The task-specific nature of the estimator performance is demonstrated through analysis, simulation, and experimentation. A strategy for selecting object priors is proposed. Acoustic scattering from kidney tissue is the emphasis of this paper, although the results are more generally applicable.