Low-dose PET image denoising based on coupled dictionary learning

Reducing the doses of tracers in positron emission tomography (PET) scans can effectively reduce the damage caused to the patient's health, but at the same time, doing so reduces the quality of the resulting image. The purpose of this study is to reduce the noise in low-dose (LD) PET images through a coupled dictionary learning method and improve the feature similarity between LD images and standard-dose images. The main method in this paper involves training a coupled dictionary with LD and standard-dose PET image samples and then utilizing the updated dictionary to reconstruct a denoised image. This article changes the dictionary training process by not directly updating the joint dictionary, but rather updating the two dictionaries separately in a specific manner. We compare a variety of traditional denoising algorithms that are applied to LD PET images. The proposed method significantly improves the quality of the resulting image, effectively removing the noise from the image and better retaining the image details. In addition, the reconstructed image is closest to the standard-dose PET image.