Deep denoising in phase contrast breast computed tomography

Abstract

Breast cancer is among the most prevalent cancers worldwide. Despite screening with digital mammography and digital breast tomosynthesis, many cancers remain undetected, and both modalities require breast compression. Breast computed tomography (BCT) is an alternative modality providing isotropic resolution without compression, but achieving sufficient image quality requires higher x-ray doses than current screening practice. Phase-contrast propagation-based computed tomography (PBCT) can provide enhanced image quality compared to absorption-based computed tomography (ABCT) at equivalent doses. In the present work, we compare supervised deep learning denoising models trained on ABCT and PBCT..