Harmonic Autoencoding Framework for Multiple Tasks in Magnetic Particle Imaging Reconstruction

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Signal Processing

Background:

• Magnetic Particle Imaging (MPI) offers high resolution for magnetic particle distribution reconstruction.
• Traditional MPI methods (X-space, system matrix) rely on accurate harmonic components, which are degraded by noise and system matrix collection challenges.
• This degradation impacts reconstruction fidelity in MPI.

Purpose of the Study:

• To propose a unified framework leveraging harmonic knowledge to enhance MPI reconstruction quality.
• To address limitations in harmonic accuracy caused by noise and system matrix complexities.
• To improve the fidelity of MPI image reconstruction.

Main Methods:

• Developed a unified framework based on harmonic knowledge for MPI reconstruction.
• Utilized autoencoders pretrained by restoring masked system matrices (SMs) to model harmonic relationships.
• Transferred pretrained encoders for spectrum denoising and SM super-resolution tasks.

Main Results:

• The proposed framework consistently outperformed state-of-the-art (SOTA) methods on spectrum denoising and SM super-resolution tasks.
• Demonstrated effectiveness using both simulated and publicly available MPI datasets.
• Achieved marked improvements in reconstruction quality for both time and frequency domain MPI data.

Conclusions:

• The harmonic knowledge-based framework effectively enhances MPI reconstruction quality.
• The approach successfully tackles challenges related to noise and system matrix complexities.
• This method represents a significant advancement for improving MPI image fidelity and performance.