| Image quality for ultrasonic imaging should be defined in terms of an observer's ability to perform a diagnostic task. In this dissertation we consider the task of lesion detection. Many factors contribute to an observer's performance, including the system design, processing algorithms, display technology, and the eye-brain system. The channel RF is the most fundamental form of data, and contains the most information of the processing, display, and observation chain. We want the information in the echo signals at the detector level to be as informative as possible with respect to detection tasks. Our approach to system design is to maximize the diagnostic information at the detector and beamformation level by maximizing the ideal observer performance. The ideal observer is the decision-maker that has the best possible performance of all observers. In particular, it has the best-possible true positive probability for all false-positive levels. An ideal observer model for ultrasound detection performance is developed by modeling essential features of objects and imaging systems using a linear systems approach. Performance metrics for ultrasonic detection tasks are developed under a low contrast approximation. The analysis provides a way of assessing and maximizing the diagnostic performance of ultrasound systems from an information content perspective. The theory reveals that curved wavefronts that occur in the nearfield of ultrasound beams can be viewed as spatial codes that can be decoded to obtain high quality images. This method is implemented on a commercial scanner, and shown to exhibit performance superior to state-of-the-art dynamic receive focusing methods. |
