Denoising with Graphics Processing Units and Deep Learning in Non-invasive Medical Imaging

Abstract

Medical imaging is not only essential to the diagnostic process, but also plays a very important role in determining the course and effectiveness of treatment. In the last few decades, tremendous technological innovations have been made in the field of non-invasive medical imaging. Among them, imaging methods represented by computed tomography and magnetic resonance imaging are indispensable in current clinical medicine because they can acquire biological structures and functions in three to four dimensions with high spatial resolution non-invasively. However, the acquisition of data with high spatial resolution generally leads to a decrease in the signal-to-noise ratio. A longer acquisition time is required to improve the signal-to-noise ratio. However, for non-invasive medical image acquisition in clinical settings, a long acquisition time is impractical and results in a decrease in signal-to-noise ratio, especially in high spatial resolution images. It is thus essential to develop effective denoising techniques as post-processing and also to adapt the optimal denoising method in accordance with the user’s objectives. This review provides a brief overview of denoising techniques as post-processing for medical imaging, and introduces our work on fast and accurate denoising methods using graphics processing units and denoising with deep learning.