Anℓ1-plug-and-play approach for MPI using a zero shot denoiser with evaluation on the 3D open MPI dataset

Area of Science:

• Medical Imaging
• Computational Imaging
• Biomedical Engineering

Background:

• Magnetic Particle Imaging (MPI) is an emerging medical imaging modality known for high temporal resolution and lack of ionizing radiation.
• MPI reconstruction from measured signals is an ill-posed problem, necessitating regularization techniques.
• Existing regularization methods include Tikhonov, iterative approaches, and machine learning, with Plug-and-Play (PnP) methods showing promise.

Purpose of the Study:

• To propose a novel Plug-and-Play (PnP) reconstruction method for MPI utilizing a generic zero-shot denoiser.
• To validate and optimize the reconstruction parameters of the proposed PnP method.
• To evaluate the performance of the zero-shot PnP approach against existing methods on a 3D MPI dataset.

Main Methods:

• A Plug-and-Play (PnP) reconstruction framework was developed, incorporating a generic zero-shot denoiser with an ℓ1-prior.
• Reconstruction parameters were optimized through an extensive search on a custom hybrid validation dataset.
• The method was applied to the 3D Open MPI Dataset and compared with Tikhonov, ART, DIP, and a previously developed PnP-MPI method.

Main Results:

• The proposed zero-shot PnP method demonstrated effective MPI reconstruction with validated parameters on the 3D Open MPI Dataset.
• Quantitative and qualitative evaluations confirmed the performance of the approach.
• The method's robustness was assessed across different data preprocessing levels.

Conclusions:

• The developed zero-shot Plug-and-Play method offers a cost-effective and efficient approach to MPI reconstruction by eliminating the need for task-specific denoiser training.
• This method presents a versatile tool applicable to future MPI contexts and advancements.