BLIND COMPRESSED SENSING WITH SPARSE DICTIONARIES FOR ACCELERATED DYNAMIC MRI.

Area of Science:

• Medical Imaging
• Magnetic Resonance Imaging (MRI)
• Signal Processing

Background:

• Dynamic MRI reconstruction often uses sparse coding with fixed dictionaries.
• Jointly estimating dictionary basis and coefficients improves reconstruction quality.
• Incorporating prior information can further refine MRI data recovery.

Purpose of the Study:

• To investigate the impact of sparsity priors on learned basis functions for dynamic MRI.
• To enhance the quality of dynamic MRI reconstructions from undersampled Fourier measurements.
• To suppress noisy basis functions through constrained learning.

Main Methods:

• Modeling voxel time series as sparse linear combinations of basis functions.
• Jointly estimating dictionary basis and sparse coefficients from k-space data.
• Applying sparsity priors in pre-specified transform domains to learned basis functions.

Main Results:

• The proposed method effectively suppresses noisy basis functions.
• Sparsity priors on basis functions lead to improved MRI reconstruction quality.
• Experimental results demonstrate the effectiveness of the constrained approach.

Conclusions:

• Sparsity priors on learned basis functions offer a valuable enhancement for dynamic MRI reconstruction.
• This technique improves the fidelity and reduces artifacts in undersampled MRI data.
• The method provides a robust framework for high-quality dynamic MRI.