Accelerated dynamic MRI exploiting sparsity and low-rank structure: k-t SLR

Area of Science:

• Medical Imaging
• Biophysics
• Signal Processing

Background:

• Dynamic MRI requires high data acquisition rates.
• Under-sampling k-t space data is common to accelerate MRI scans.
• Classical methods rely on Fourier space properties, limiting performance with complex motion.

Purpose of the Study:

• To introduce a novel algorithm for reconstructing dynamic MRI data from under-sampled k-t space.
• To leverage data correlations using the Karhunen-Loeve Transform (KLT) domain for improved reconstruction.
• To address limitations of existing KLT-based methods by posing reconstruction as a spectrally regularized matrix recovery problem.

Main Methods:

• Utilized a data-dependent Karhunen-Loeve Transform (KLT) for compact data representation.
• Formulated the reconstruction as a spectrally regularized matrix recovery problem.
• Simultaneously determined temporal basis functions and spatial weights from measured data, incorporating data sparsity.

Main Results:

• Achieved high-quality dynamic MRI reconstructions at various acceleration factors.
• Demonstrated superior performance compared to existing methods using numerical phantoms.
• Validated the algorithm's effectiveness with in vivo cardiac perfusion MRI data.

Conclusions:

• The proposed spectrally regularized matrix recovery scheme offers significant improvements in dynamic MRI reconstruction.
• The KLT-based approach effectively exploits data correlations and sparsity for enhanced image quality.
• This novel method is suitable for diverse dynamic imaging scenarios, including those with non-periodic motion.