Beamforming designs for breast sonography

Abstract

The ideal observer (IO) methodology enables designers to derive optimal beamforming strategies for specific diagnostic tasks or general classes of tasks, although fast implementation requires approximations. We explore this approach in an effort to assess beamforming performance for a range of lesion-feature discrimination challenges. We show that matchedfilter (MF), linearly-constrained minimum-variance (MV) with low-rank approximation, and Wiener-filter (WF) beamformers each emerge as approximations to the ideal-observer strategy for low-contrast lesion discrimination. For high-contrast tasks, we evaluate an iterative Wiener-filter (IWF) beamformer as a computationally-intense method for minimizing contrast assumptions when prior information is available. I. INTRODUCTIONIdeal observer (IO) calculations provide designers with theoretical upper limits on task-specific visual diagnostic performance in sonography. They also suggest optimal beamforming strategies for discriminating specific lesion features. In this study, beamformers are evaluated based on their ability to transfer diagnostic information from patient to images in a way that is accessible to observers. Observer performance is measured via receiver-operating characteristic (ROC) analysis applied to an ensemble of simulated lesion images generated via Monte Carlo methods [1].We have used the IO formalism to (a) state the general objective of beamforming with respect to lesion discriminability and (b) show how different approximations to the IO strategy result in a range of beamformers discussed in the literature such as matched filter (MF), minimum variance (MV) and Wiener filter (WF) beamformer designs. From this perspective, strengths and weaknesses of each beamforming approach are compared.