Deep learning-based noise reduction method for the system matrix in magnetic particle imaging

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Artificial Intelligence

Background:

• Magnetic Particle Imaging (MPI) is an emerging technique using superparamagnetic iron oxide nanoparticles for high-sensitivity, rapid imaging.
• Noise during system matrix (SM) calibration and signal acquisition degrades MPI image quality.
• Effective noise suppression is crucial for enhancing MPI performance.

Purpose of the Study:

• To develop a deep learning-based model for efficient noise suppression in MPI.
• To improve the overall quality of MPI images by addressing noise artifacts.

Main Methods:

• A hybrid encoder-decoder network integrating residual blocks (Res-Blocks) and swin transformer modules was proposed.
• The model utilizes multi-scale feature extraction to separate noise from signals.
• Cross-level feature fusion was employed for optimized frequency-domain recovery.

Main Results:

• The denoised system matrix (SM) showed an average 12 dB improvement in signal-to-noise ratio (SNR).
• Reconstructed MPI images exhibited enhanced visual quality with a peak SNR of 29.11 dB.
• The model achieved a structural similarity index of 0.93, outperforming existing methods.

Conclusions:

• The developed deep learning model offers a robust solution for noise suppression in MPI system matrices.
• This noise reduction framework can enhance MPI image quality and is potentially extensible to other SM-based imaging modalities.