Learning Background Noise for Tabletop Magnetic Particle Imaging Reconstruction

Magnetic particle imaging is an emerging tomographic technique that utilizes the nonlinear magnetization response of superparamagnetic iron oxide nanoparticles to provide high-contrast and high-spatial-resolution images. However, image quality can be affected by various noise sources, including harmonic interference, Gaussian noise, and ambient noise, which can reduce the signal-to-noise ratio and introduce artifacts. To address these challenges, we propose a novel end-to-end neural network approach for magnetic particle imaging reconstruction, which can learn complex nonlinear relationships in training data, including complex background noise features and missing higher-order harmonic information. Unlike traditional methods, the neural network we propose can directly reconstruct images from raw signals, effectively simplifying the image reconstruction process. In addition, we have also made improvements in the hardware part, which can effectively improve the signal-to-noise ratio of the signals. The experimental results show that our method effectively enhances the quality of the reconstructed image, and significantly accelerates the reconstruction process.