Blind compressive sensing dynamic MRI.

Area of Science:

• Medical Imaging
• Signal Processing
• Optimization Algorithms

Background:

• Dynamic magnetic resonance imaging (dMRI) requires significant data acquisition time.
• Undersampled measurements in dMRI lead to challenges in image reconstruction.
• Existing compressed sensing methods often rely on predefined dictionaries or low-rank assumptions.

Purpose of the Study:

• To develop a novel blind compressive sensing (BCS) framework for recovering dynamic MRI data from undersampled measurements.
• To simultaneously estimate the temporal basis dictionary and sparse coefficients from undersampled data.
• To improve reconstruction quality and acceleration rates in dynamic MRI.

Main Methods:

• Modeling dynamic MRI signals as sparse linear combinations of temporal basis functions from an estimated dictionary.
• Formulating reconstruction as a constrained optimization problem with data consistency and L1-norm sparsity priors.
• Employing a majorize-minimize algorithm with alternating minimization for efficient computation.
• Utilizing a Frobenius norm dictionary constraint to avoid scale ambiguity.

Main Results:

• The BCS framework achieves improved reconstructions at high acceleration rates compared to low-rank methods.
• The proposed algorithm is significantly faster than existing sparse coding and dictionary estimation approaches.
• BCS demonstrates better recovery rates than classical Fourier-based CS and is robust to local minima.
• Superior reconstruction performance is observed for contrast-enhanced dynamic MRI data.

Conclusions:

• The novel BCS framework offers a powerful tool for accelerating dynamic MRI acquisition.
• Simultaneous dictionary and coefficient estimation from undersampled data enables high-quality dMRI reconstruction.
• The method is particularly beneficial for dynamic MRI applications where signals lack sparsity in known dictionaries.