Robust GRAPPA reconstruction using sparse multi-kernel learning with least squares support vector regression.

Area of Science:

• Magnetic Resonance Imaging (MRI)
• Medical Imaging
• Image Reconstruction

Background:

• Accurate interpolation coefficient fitting is vital for k-space-based parallel MRI reconstruction.
• Conventional generalized autocalibrating partially parallel acquisitions (GRAPPA) and nonlinear GRAPPA (NLGRAPPA) are sensitive to interpolation window and kernel parameters, limiting reconstruction consistency.
• Existing methods struggle to consistently achieve optimal results across various acceleration factors and coil configurations.

Purpose of the Study:

• To develop a robust parallel MRI reconstruction method using sparse multi-kernel learning.
• To adaptively determine interpolation coefficients and kernel weights for improved image reconstruction.
• To reduce sensitivity to interpolation window and kernel parameters compared to NLGRAPPA.

Main Methods:

• Sparse multi-kernel learning within a least squares support vector regression framework.
• Adaptive determination of kernel combination weights and interpolation coefficients using semi-infinite linear programming.
• Reconstruction of images under diverse subsampling patterns and coil datasets.

Main Results:

• The proposed method achieves an optimized balance between noise suppression and artifact reduction for various sampling schemes.
• Experimental results on phantom and in vivo data demonstrate robust image reconstruction.
• The method shows significantly reduced sensitivity to interpolation window and kernel parameters compared to NLGRAPPA.

Conclusions:

• Sparse multi-kernel learning offers a robust and adaptive approach for parallel MRI reconstruction.
• The developed method enhances image quality and consistency across different acquisition parameters.
• This technique provides a more reliable alternative to existing GRAPPA-based reconstruction methods.