Scan-Adaptive MRI Undersampling Using Neighbor-based Optimization (SUNO).

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Computational Imaging

Background:

• Accelerated MRI reduces scan time through efficient data acquisition or advanced reconstruction.
• Undersampling with learning-based reconstruction is a key acceleration strategy.
• Population-adaptive sampling patterns improve MRI design but may miss individual scan details.

Purpose of the Study:

• To develop a framework for jointly learning scan-adaptive undersampling patterns and reconstruction models for personalized MRI.
• To enable tailored MRI sampling for individual scans, capturing subject- or slice-specific details.
• To improve the efficiency and image quality of accelerated MRI.

Main Methods:

• Proposed a framework (SUNO) for jointly learning scan-adaptive Cartesian undersampling patterns and reconstruction models.
• Utilized an alternating algorithm with iterative coordinate descent (ICD) for offline optimization of sampling patterns.
• Employed a nearest neighbor search for selecting scan-adaptive patterns at test time based on low-frequency k-space data.

Main Results:

• Demonstrated improved performance on the fastMRI multi-coil knee and brain datasets.
• Achieved better visual quality and quantitative metrics compared to existing undersampling patterns at 4x and 8x acceleration factors.
• The SUNO framework enables more tailored sampling for individual MRI scans.

Conclusions:

• The SUNO framework effectively learns personalized undersampling patterns and reconstruction models for accelerated MRI.
• This approach enhances image quality and efficiency in MRI, outperforming generic methods.
• The developed framework offers a promising direction for personalized medical imaging acquisition.