Frequency Structure Consistency Learning Approach for Fast Signal Matrix Measurement in Magnetic Particle Imaging

A core challenge for accelerating high-resolution image reconstruction in magnetic particle imaging (MPI) is to rapidly measure the high-resolution signal matrix. For accurately measuring the high-resolution signal matrix, the number of meticulous measurements at numerous positions in the field-of-view proves time-consuming (measurement of a 37 × 37 × 37 signal matrix takes about 32 hours). To improve reconstructed image quality and shorten signal matrix measurement time, existing methods explore deep learning methods to use time-saving measured low-resolution signal matrix (a 9 × 9 × 9 signal matrix just takes about 0.5 hours) generating high-resolution signal matrix. However, previous methods show poor performance for high-frequency signal recovery in signal matrix, which is crucial for reconstructing high-resolution image. In this work, to achieve a high-resolution signal matrix recovery and shorten its acquisition time, we propose a novel learning approach of frequency structure consistency loss and signal component embedding strategy. Our approach guides model to learn the distribution of frequency of signal matrix. We evaluate our methods and state-of-the-art methods on the two simulation datasets and four public measured signal matrices in Open MPI Data. The results show that our method outperforms the compared methods in high-frequency structural signal recovery. Additionally, our method can recover a high-resolution signal matrix based on a down-sampling factor of 16 less than 15 seconds, which accelerates the acquisition time over 60 times faster than the measurement-based high-resolution signal matrix with the minimum error (nRMSE = 0.041).