Wiener Filter for Filtered Back Projection in Digital Breast Tomosynthesis

Conventional filtered back projection (FBP) reconstruction for digital breast tomosynthesis (DBT) can suffer from a low signal to noise ratio. Because of the strong amplification by the reconstruction filters (ramp, apodization and slice thickness), noise at high spatial frequencies can be greatly increased. Image enhancement by Wiener filtering is investigated as a possible method to improve image quality. A neighborhood wavelet coefficient window technique is used to estimate the noise content of projection images and a Wiener filter is applied to the projection images. The neighborhood wavelet coefficient window is a non-linear technique, which may cause the Wiener filters estimated before and after the application of the reconstruction filters to be different. Image quality of a FBP reconstruction with and without Wiener filtering is investigated using a Fourier-based observer detectability metric (d') for evaluation. Simulations of tomosynthesis are performed in both homogeneous and anatomic textured backgrounds containing low-contrast masses or small microcalcifications. Initial results suggest that improvements in detectability can be achieved when the Wiener filter is applied, especially when the Wiener filter is estimated for the reconstruction filtered projections.