Comparison and Optimization of Iterative Reconstruction Algorithms in Digital Breast Tomosynthesis

Abstract Breast cancer has the highest morbidity and mortality among women. Early detection and treatment can effectively improve survival. The digital breast tomosynthesis (DBT) technique uses reconstruction algorithms to get a three-dimensional (3D) image set from a plurality of projection images that are obtained at limited view angles. In this paper, through computer simulation, out-of-focus artifact spread function (ASF), noise spectrum function, and modulation transfer function (MTF) were evaluated to optimize three iterative reconstruction algorithms, including simultaneous algebraic reconstruction technique (SART), maximum-likelihood expectation maximization (MLEM), and ordered-subset MLEM (OS-MLEM). The projection data of phantom were simulated and obtained on Geant4 Application for Tomographic Emission (GATE) platform. Based on the elimination of out-of-focus plane artifacts, the suppression of noise, and the sharpness of the reconstructed objects in frequency domain, our results suggested that the overall best reconstructed image quality was achieved when the iteration numbers were 4, 6, and 3 for SART, MLEM and OS-MLEM, respectively.