Multi‐task generative model for high‐quality magnetic particle imaging reconstruction

BACKGROUND: Magnetic Particle Imaging (MPI) is an emerging technology used to visualize the motion of magnetic nanoparticles in biological tissues. Due to the complexity of the physical behavior of nanoparticles, it is challenging to reconstruct high-quality MPI images from MPI signals. Traditional reconstruction methods, such as system matrix and x-space, are either extremely time-consuming or result in very blurry images. Recently, generative models have been employed for MPI image reconstruction and have achieved success. However, these models still have limitations in obtaining high-quality MPI images. PURPOSE: Here, we propose a novel multi-task generative method to realize high-quality MPI image reconstruction. METHODS: In this method, the generative model simultaneously undertakes two MPI image processing tasks: reconstruction and segmentation. The main task of image reconstruction generates MPI images, while the auxiliary task of image segmentation guides the main task to focus on key areas of objects in MPI images during the generation process. RESULTS: Our experimental results showed that the proposed multi-task model, with its superior generalization ability, outperforms both traditional MPI reconstruction methods and single-task generative methods. Our results also suggested that the tasks of image generation and image segmentation significantly promote each other during the MPI image reconstruction. CONCLUSION: In our current study, we propose a deep-learning based multi-task method to realize high-quality MPI (Magnetic Particle Imaging) image reconstruction. As far as we know, it is the first attempt to perform feature sharing on MPI image reconstruction and segmentation in medical image analysis.