Predicting target detectability in acoustic elastography

The clinically relevant task of visually detecting low-contrast targets in noisy strain images is studied. Detectability is measured quantitatively using contrast-to-noise ratio (CNR) analysis. Contrast in strain images is generated by a complex interaction between the soft tissue elasticity shear modulus distribution, target shape and location in the stress field, and external boundary conditions. While large strain variation is preferred for enhancing the contrast, this also increases the noise in strain estimates in a nonlinear fashion. Therefore, understanding the tradeoffs between contrast and noise is necessary for improving the diagnostic performance of strain imaging. In this paper, targets having slab, cylindrical and spherical geometries are studied. Strains in the target and background and the precision of their estimates are described in terms of the corresponding shear modulus values for each geometry. These results are then incorporated into the CNR expression to investigate the changes in target detectability with the shear modulus variation, signal parameters (echo signal-to-noise ratio (SNR_i) and fractional bandwidth Y), and signal processing variables (time-bandwidth product W and fractional window overlap ΔT˜T). The results include: 1) formulas describing target and background strains for the three geometries as a function of the applied compression, boundary conditions and shear modulus values; 2) mathematical description of the consequences that nonuniformities in tissue elasticity and variations in strain contrast with the target geometry impose upon detectability; and 3) demonstration of the need to carefully choose values for signal processing variables