The Generalized Contrast-to-Noise Ratio

摘要

Many adaptive algorithms claim to provide higher contrast than delay-and-sum (DAS). These claims are often backed by estimations of the contrast-to-noise ratio (CNR). Intuitively, we assume that higher CNR leads to higher probability of lesion detection, and this is indeed the case for DAS. However, non-linear processing can arbitrarily alter CNR, and yet yield no improvement in the detection probability. We propose a new image quality index, the generalized contrast-to-noise ratio (GCNR), based on the overlap area of the probability density function inside and outside the target area. GCNR can be used with non-linear beamforming algorithms, remaining unaltered if the dynamic range is changed. We demonstrate that GCNR is proportional to the maximum success rate that can be expected from the algorithm. Using Field II, we compare the performance of CNR and GCNR in 6 imaging algorithms. While CNR varies significantly between the 6 algorithms, we do not observe notable variations in GCNR (<;10%), which means that the 6 algorithms have similar lesion detection capabilities. GCNR fixes the methodological flaw of using CNR with algorithms that alter the probability density function of the ultrasound signal, and allows us to assess the significance of contrast enhancing effects in ultrasound imaging.