Performance evaluation of digital breast tomosynthesis systems: comparison of current virtual clinical trial methods

Abstract

Virtual clinical trials (VCT) have been developed by a number of groups to study breast imaging applications, with the focus on digital breast tomosynthesis imaging. In this review, the main components of these simulation platforms are compared, along with the validation steps, a number of practical applications and some of the limitations associated with this method. VCT platforms simulate, up to a certain level of detail, the main components of the imaging chain: the x-ray beam, system geometry including the antiscatter grid and the x-ray detector. In building VCT platforms, groups use a number of techniques, including x-ray spectrum modelling, Monte Carlo simulation for x-ray imaging and scatter estimation, ray tracing, breast phantom models and modelling of the detector. The incorporation of different anthropomorphic breast models is described, together with the lesions needed to simulate clinical studies and to study detection performance. A step by step comparison highlights the need for transparency when describing the simulation frameworks. Current simulation bottlenecks include resolution and memory constraints when generating high resolution breast phantoms, difficulties in accessing/applying relevant, vendor specific image processing and reconstruction methods, while the imaging tasks considered are generally detection tasks without search, evaluated by computational observers. A number of applications are described along with some future avenues for research.