Self-supervised Signal Denoising for Magnetic Particle Imaging

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Signal Processing

Background:

• Magnetic Particle Imaging (MPI) offers high-resolution, high-sensitivity tracking of superparamagnetic iron oxide nanoparticles (SPIONs).
• Signal distortions from various noises degrade MPI imaging quality.
• Existing threshold-based methods are ineffective against dynamic noise in MPI signals.

Purpose of the Study:

• To introduce a self-supervised denoising method for improving MPI signal quality.
• To address the limitations of traditional denoising techniques in MPI.
• To enhance the overall image quality of MPI by reducing noise.

Main Methods:

• A U-net based deep learning architecture was adapted for MPI signal denoising.
• The network was trained using two periods of noisy MPI signals.
• Shape prior knowledge of MPI signals was incorporated to improve self-supervised network convergence.

Main Results:

• The proposed learning-based method successfully denoises MPI signals without requiring labeled data.
• The method demonstrates improved image quality compared to traditional techniques.
• The approach achieved superior performance over other self-supervised methods in MPI signal denoising.

Conclusions:

• Self-supervised learning offers a viable solution for denoising MPI signals, even without labeled datasets.
• The developed method effectively enhances MPI image quality by mitigating dynamic noise.
• This technique represents a significant advancement for MPI signal processing and applications.