Nonlinear GRAPPA: a kernel approach to parallel MRI reconstruction.

Area of Science:

• Medical Imaging
• Machine Learning in Radiology

Background:

• Grappa (Generalized Autocalibrating Partially Parallel Acquisitions) is a common MRI reconstruction technique.
• High acceleration factors in Grappa can lead to increased noise and reduced image quality, especially with limited auto-calibration signals (ACS).

Purpose of the Study:

• To develop and evaluate a nonlinear method for improving Grappa-based MRI reconstruction.
• To address the limitations of conventional Grappa at high acceleration factors.

Main Methods:

• A kernel-based nonlinear transformation maps undersampled k-space data to a high-dimensional feature space.
• Linear combination coefficients are derived from ACS data in this feature space, effectively adding virtual channels.
• A polynomial kernel with explicit mapping functions was investigated.

Main Results:

• The proposed nonlinear Grappa method demonstrated significant improvements in reconstruction quality.
• Experimental results on phantom and in vivo data showed superior performance over standard Grappa and other advanced methods.
• Reduced noise levels were observed even at high acceleration factors.

Conclusions:

• Nonlinear kernel methods offer a powerful approach to enhance MRI reconstruction beyond linear techniques like Grappa.
• This nonlinear Grappa method provides a substantial improvement in image quality for accelerated MRI acquisition.