ABSTRACT
Imaging systems are evaluated from objective assessments that relate ultimately to observer performance. The observer can be an expert human or an algorithm evaluating criteria based on decision theory [2]. Prominent among the latter is the Bayesian ideal observer-often referred to simply as the ideal observer-that combines all available information to make the decision, and thus it achieves optimal task performance [1]. If the performance measured by a panel of expert radiologists is significantly less than the ideal, the system should be redesigned, but only if it is determined that the acquisition stage of image formation (including output power, noise, transducer properties, and beamforming aspects) is limiting performance. Sometimes task information is present in the image but difficult to observe; for example, flowing-blood echoes are found in recorded echo signal but are difficult to see without Doppler processing and color overlays. When the display stage limits performance, image processing is often very helpful.
