Improving image reconstruction for digital breast tomosynthesis

Abstract:

Digital breast tomosynthesis (DBT) is an advanced form of breast imaging that uses a low-dose x-ray system and tomographic reconstruction to create 3-D images of the breast for breast cancer screening and diagnosis. Studies have shown that DBT can reduce both false-negative diagnoses and false-positive recalls compared to mammography alone. My PhD research focuses on improving the image quality for DBT. Statistical image reconstruction or model-based image reconstruction (MBIR) have shown great success in achieving superior reconstruction quality and lowering x-ray dose in different 3D modalities. Inspired by the idea of statistical image reconstruction or MBIR, my PhD study demonstrated the improvement of DBT image quality with the following approaches: (1) improving the accuracy of digital projector with a more realistic detection model and applying it to subpixel image reconstruction; (2) incorporating detector blur in the forward model and the noise correlation in the optimization problem; (3) removing truncated projection artifacts by extrapolating the projection views or using regularized reconstruction; (4) estimating the shift-variance of the source blur for the DBT geometry and its effect on the spatial resolution. The results of these studies indicate the great potential of MBIR to improve DBT reconstruction.