Analysis of minimum-variance and Wiener-filtered beamforming strategies

We show that the minimum-variance (MV),Wiener-filtered (WF), and other beamformers can be derived as approximations to the ideal-observer´s strategy for lesion feature discrimination. We analyze breast lesion discrimination performance for five beamformers. Four of the five include matching filtering of receive-channel signal before summation, because there is no loss of task information in the RF signals due to beamforming. Differences among beamformers occur depending on how they prepare RF signals for demodulation. Generally the MV and WF beamformers decorrelate the echo signals the most before demodulation, which transfers more task information into the B-mode data. We find the WF beamformer performs better than others because it is the closest approximation to the ideal observer strategy. The MV beamformer requires an additional low-rank approximation that handicaps performance for discriminating four of the five lesion features, but performs well for the fifth task because reducing rank filters data that is well matched to the task 5 feature spectrum. However, similarly reducing the rank of the WF filter makes it the beamformer of choice.