DDMPI: Diffusion Denoising for Magnetic Particle Imaging at the Low Concentration

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Magnetic Particle Imaging (MPI) offers high sensitivity and temporal resolution for preclinical studies.
• MPI images often suffer from stripe artifacts, especially with low concentrations of magnetic particles used in biological applications.
• These artifacts hinder the identification of crucial details in MPI scans.

Purpose of the Study:

• To develop a novel method for artifact removal in low-concentration MPI images.
• To improve the diagnostic utility of MPI by enhancing image quality.
• To create a versatile solution applicable to various MPI devices and imaging scenarios.

Main Methods:

• Proposed a joint optimization approach named Diffusion Denoising Model for MPI (DDMPI).
• Integrated a diffusion model with a Transformer architecture for direct artifact removal from MPI images.
• Utilized a latent encoder, conditional latent diffusion model, and a U-shape Transformer with strip-self-attention.

Main Results:

• DDMPI effectively removed stripe artifacts from MPI images in low-concentration scenarios.
• The model successfully recovered fine details, significantly improving image quality.
• Ablation studies confirmed the effectiveness of individual modules within DDMPI.
• Extensive experiments (simulation, phantom, in vivo) validated the performance.

Conclusions:

• DDMPI offers a robust solution for artifact reduction in MPI, particularly for low-concentration biological applications.
• The method is independent of primary image reconstruction techniques, allowing flexible integration with existing MPI devices.
• DDMPI has the potential to significantly advance in vivo MPI imaging and disease diagnosis.