Deep-learning-based denoising network for reducing MPI multiple repetition measurements

Abstract

Magnetic particle imaging (MPI) is an emerging medical imaging technique with high temporal resolution, and has the potential for real-time imaging. However, current MPI systems usually require multi-repetition measurement averaging for signal denoising, which diminishes the temporal resolution of MPI. Therefore, in this study, we use the deep-learning method to resolve the problem. Specifically, MPI images with less repetition times are used to predict the high-quality images with more repetition times using the neural network (e.g., UNet). In such cases, high-quality images can still be obtained with high temporal resolution. Our method is evaluated on the simulation dataset and shows its superiority. We will further evaluate the method in the real-world dataset and discuss its applicability.