Low-Rank and Adaptive Sparse Signal (LASSI) Models for Highly Accelerated Dynamic Imaging

Area of Science:

• Medical Imaging
• Image Processing
• Signal Recovery

Background:

• Sparsity and low-rank techniques are effective for inverse problems in dynamic imaging.
• Existing methods like L+S and dictionary-blind compressed sensing have shown promise but can be improved.
• Dynamic imaging often requires reconstruction from undersampled measurements.

Purpose of the Study:

• To introduce LASSI, a data-adaptive extension of the L+S model for dynamic imaging.
• To develop efficient methods for jointly estimating dynamic signal components and adaptive dictionaries.
• To demonstrate the effectiveness of LASSI for dynamic magnetic resonance image reconstruction.

Main Methods:

• Decomposition of temporal image sequences into low-rank and sparse spatiotemporal (3D) patches.
• Utilizing adaptive dictionary learning for sparse representation.
• Joint estimation of dynamic signal components and the spatiotemporal dictionary.
• Formulating sparsity-penalized dictionary-blind compressed sensing as a special case.

Main Results:

• LASSI schemes show promising performance in dynamic magnetic resonance image reconstruction.
• LASSI outperforms existing methods like k-t SLR and L+S.
• LASSI achieves competitive results compared to dictionary-blind compressed sensing methods.

Conclusions:

• LASSI offers a powerful and data-adaptive approach for dynamic imaging reconstruction.
• The method effectively leverages sparsity and low-rank properties in an adaptive dictionary domain.
• LASSI advances the field of compressed sensing for dynamic imaging applications.