Denoising digital breast tomosynthesis projections using deep learning with synthetic data as training set.

We propose a solution for the lack of suitable training data to train DNN models for denoising DBT projections, showing that we just need the synthesized noise to be in the same sample space as the target image.