SPARSITY-DRIVEN PARALLEL IMAGING CONSISTENCY FOR IMPROVED SELF-SUPERVISED MRI RECONSTRUCTION.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Physics-driven deep learning (PD-DL) models enhance rapid MRI reconstruction.
• Self-supervised learning is used when fully-sampled data is unavailable.
• High acceleration rates in MRI can cause artifacts, reducing image quality.

Purpose of the Study:

• To develop a novel training strategy for PD-DL networks to improve MRI reconstruction.
• To mitigate artifacts and noise amplification in accelerated MRI scans.
• To enhance image fidelity in self-supervised learning for MRI.

Main Methods:

• Proposed a new training approach for PD-DL networks using designed perturbations.
• Enhanced k-space masking with a novel consistency term for perturbation prediction.
• Assessed model's ability to predict perturbations in a sparse domain.

Main Results:

• The novel training strategy effectively reduced aliasing artifacts in MRI.
• Noise amplification was mitigated at high acceleration rates.
• Outperformed state-of-the-art self-supervised methods on fastMRI knee and brain datasets.

Conclusions:

• The proposed method enables reliable, artifact-free MRI reconstructions.
• This approach improves the performance of self-supervised learning for accelerated MRI.
• Offers a promising solution for high-quality, rapid MRI acquisition.