Physics-informed Machine Learning Approach for Denoising Low Resolution Diffusion Magnetic Resonance Images

Abstract

Image denoising is to remove noise from a noisy image for the purpose of restoration to the true image. Meanwhile, since noise, edge and texture are high frequency components, it is difficult to distinguish them in during denoising and the denoised images could inevitably lose some details. However, recovering meaningful information from noisy images in the process of noise removal to obtain high quality images is an important research problem in recent time. Although, image denoising is a classic problem which has been subject to studies for a long time, it remains a challenging and open task. This is because from a mathematical perspective, image denoising is an inverse problem with a non-unique solution. In recent time, various deep learning methods have been adopted to address denoising problem. Therefore, there have been efforts to develop machine learning methods with Physics contraints. This study attempts developing a Physics-informed machine learning model in which a modified Bloch-Torrey formulation of diffusion signal is used as a constraint.