DEQ-MPI: A Deep Equilibrium Reconstruction With Learned Consistency for Magnetic Particle Imaging

Area of Science:

• Biomedical Imaging
• Medical Physics
• Machine Learning

Background:

• Magnetic Particle Imaging (MPI) provides high-contrast imaging for magnetic nanoparticles.
• Image reconstruction in MPI is an ill-posed problem requiring data consistency and regularization.
• Existing learned priors for MPI reconstruction face challenges with inference time and generalization.

Purpose of the Study:

• To develop a novel physics-driven MPI reconstruction method using a deep equilibrium model.
• To improve image quality and reduce inference time compared to current state-of-the-art methods.
• To address limitations of traditional hand-crafted priors and conventional unrolling methods.

Main Methods:

• Introduced DEQ-MPI, a deep equilibrium model with learned data consistency for MPI reconstruction.
• Augmented neural networks into iterative optimization, inspired by deep learning unrolling methods.
• Trained an implicit mapping for convergent solutions and incorporated a learned consistency measure.

Main Results:

• DEQ-MPI demonstrated superior image quality on simulated and experimental data.
• Achieved competitive inference times compared to existing MPI reconstruction techniques.
• Showcased improved data distribution capture through learned consistency measures.

Conclusions:

• DEQ-MPI represents a significant advancement in MPI image reconstruction.
• The method offers a promising approach for high-quality and efficient MPI imaging.
• Physics-driven deep learning models can effectively address challenges in complex imaging modalities like MPI.