Dynamic Sparse Mask Spatial-Frequency System Matrix Compression for Accelerated Reconstruction in Narrowband Magnetic Particle Imaging.

OBJECTIVE

Narrowband magnetic particle imaging (MPI), owing to its inherently high signal-to-noise ratio (SNR), has shown strong potential for enabling large-field-of-view (FOV) scanners suitable for human-scale imaging. System-matrix (SM)-based reconstruction can effectively suppress image blurring and negative artifacts. However, enlarging the FOV greatly increases the SM size, leading to prohibitive memory usage and prolonged reconstruction time.

METHODS

To address this challenge, we propose a Spatial-Frequency System Matrix (SF-SM) approach that exploits the sparse representation of calibration responses in the discrete cosine transform (DCT) domain to achieve substantial matrix reduction. Subsequently, a Dynamic Sparse Mask (DSM) strategy is introduced to retain the most informative spatial-frequency components: a global prior mask removes consistently low-contribution frequency components, while a signal-driven mask derived from the measured harmonic image captures object-dependent spectral characteristics. Their intersection serves to extract a task-adaptive specific SM that retains informative components and suppresses noise-dominated high-frequency terms.

RESULTS

Extensive validation using simulation studies, a public dataset, and an in-house narrowband MPI scanner demonstrates that the proposed approach compresses the SM to approximately 1% of its original size in the best cases, accelerates reconstruction by nearly two orders of magnitude, and maintains or even enhances image quality relative to the Raw-SM.

CONCLUSION

The proposed DSM-based SF-SM enables substantial SM compression and highly efficient reconstruction while preserving image quality.

SIGNIFICANCE: DSM-based SF-SM provides an efficient and scalable solution for accelerating narrowband MPI reconstruction, particularly in large-FOV systems and applications requiring real-time imaging performance.