Coil sketching for computationally efficient MR iterative reconstruction.

Area of Science:

• Medical Imaging
• Computational Science

Background:

• Parallel imaging and compressed sensing in MRI face computational challenges, particularly for 3D non-Cartesian acquisitions.
• Coil compression, a common solution, reduces computational cost but introduces artifacts by losing signal energy, compromising image quality in 3D non-Cartesian imaging.

Purpose of the Study:

• To introduce coil sketching, a novel and adaptable method for computationally efficient iterative MRI image reconstruction.
• To address the limitations of coil compression in 3D non-Cartesian MRI by preserving signal energy.

Main Methods:

• The method adapts randomized sketching algorithms from machine learning and big data analysis for MRI reconstruction.
• A structured sketching matrix is employed, incorporating high-energy virtual coils (via PCA) and random combinations of low-energy coils to leverage information from all channels.

Main Results:

• Ablation experiments validated the sketching matrix design, showing improved computational efficiency and preserved signal-to-noise ratio (SNR) on both Cartesian and non-Cartesian datasets.
• On high-dimensional 3D cones datasets, coil sketching achieved up to threefold faster reconstructions with equivalent image quality compared to existing methods.

Conclusions:

• Coil sketching offers a general and versatile framework for MRI reconstruction.
• The method enables computationally fast and memory-efficient image reconstruction, particularly beneficial for large and complex datasets.