Spatial signal distribution learning for high-resolution 3D system matrix calibration in magnetic particle imaging

Area of Science:

• Magnetic Particle Imaging (MPI)
• Image Reconstruction
• Medical Imaging Technology

Background:

• High-resolution system matrices (HR-SMs) are crucial for accurate image reconstruction in 3D Magnetic Particle Imaging (MPI).
• Traditional methods for obtaining HR-SMs are time-consuming and expensive, limiting their practical application.

Purpose of the Study:

• To develop an efficient, learning-based method for 3D HR-SM calibration in MPI.
• To reduce the workload associated with acquiring HR-SMs while preserving reconstruction accuracy.

Main Methods:

• A novel spatial signal distribution learning method using a channel-decoupled multi-path state space model (MPC-SSM) was developed.
• The 3D SM is transformed into complementary spatial sequences, utilizing diverse traversal paths to capture complex spatial dependencies.
• Feature channels are grouped and assigned to path-specific SSMs for efficient modeling and reduced computational cost.

Main Results:

• The MPC-SSM method demonstrated lower normalized reconstruction error compared to existing interpolation and deep learning techniques on both simulated and real MPI data.
• Evaluations under 2x and 4x upsampling showed improved downstream image reconstruction quality.
• The method was validated on datasets including OpenMPI.

Conclusions:

• The proposed MPC-SSM offers a scalable and practical solution for 3D HR-SM calibration in MPI.
• This approach provides a generalizable modeling strategy for structured 3D medical data, enhancing the efficiency and accuracy of MPI systems.