Learned k-space Partitioning for Optimized Self-Supervised MRI Reconstruction.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Self-supervised magnetic resonance imaging (MRI) reconstruction methods train deep learning networks without fully sampled reference data.
• Existing self-supervised via data under-sampling (SSDU) methods use heuristic k-space partitioning, leading to suboptimal performance and inflexibility with changing under-sampling patterns.

Purpose of the Study:

• To develop a novel approach for learning optimal k-space partitioning in self-supervised MRI reconstruction.
• To improve the performance and adaptability of self-supervised MRI reconstruction methods.

Main Methods:

• Proposed a novel approach to learn optimal k-space partitioning by modeling a probability distribution using the LOUPE framework.
• Introduced a weighted dual-domain self-supervised loss function incorporating both k-space and image-space loss terms.

Main Results:

• The proposed dual-domain learned partitioning method outperforms existing partitioning strategies on the fastMRI dataset.
• The method adapts to new sampling patterns without requiring hand-picked partitioning methods.

Conclusions:

• Learned k-space partitioning offers superior performance and adaptability for self-supervised MRI reconstruction.
• This approach can be directly applied to under-sampled clinical MRI data, eliminating the need for fully sampled datasets.