A mathematical framework for including various sources of variability in a task-based assessment of digital breast tomosynthesis

Abstract

For a rigorous x-ray imaging system optimization and evaluation, the need for exploring a large space of many different system parameters is immense. However, due to the high dimensionality of the problem, it is often infeasible to evaluate many system parameters in a laboratory setting. Therefore, it is useful to utilize computer simulation tools and analytical methods to narrow down to a much smaller space of system parameters and then validate the chosen optimal parameters by laboratory measurements. One great advantage of using the simulation and analytical methods is that the impact of various sources of variability on the system's diagnostic performance can be studied separately and collectively. Previously, we have demonstrated how to separate and analyze noise sources using covariance decomposition in a task-based approach to the assessment of digital breast tomosynthesis (DBT) systems in the absence of x-ray scatter and detector blur. 1, 2 In this work, we analytically extend the previous work to include x-ray scatter and detector blur. With use of computer simulation, we also investigate the use of the convolution method for approximating the scatter images of structured phantoms in comparison to those computed via Monte Carlo. The extended method is comprehensive and can be used both for exploring a large parameter space in simulation and for validating optimal parameters, chosen from a simulation study, with laboratory measurements.