Dynamic cardiac MRI reconstruction using motion aligned locally low rank tensor (MALLRT)

Area of Science:

• Medical Imaging
• Biomedical Engineering
• Signal Processing

Background:

• Compressed sensing dynamic cardiac MRI relies on sparse transform models.
• Existing low-rank tensor models (Tucker decomposition) can oversmooth local details and are sensitive to motion artifacts.
• Global tensor modeling struggles with spatiotemporal correlations corrupted by frame misalignment.

Purpose of the Study:

• To present a novel Motion Aligned Locally Low Rank Tensor (MALLRT) model for dynamic MRI reconstruction.
• To overcome limitations of global tensor models, specifically oversmoothing and motion sensitivity.
• To improve image reconstruction quality in undersampled dynamic cardiac MRI.

Main Methods:

• Developed the MALLRT model enforcing low-rank constraints on image patch-based local tensors.
• Utilized group-wise inter-frame motion registration for accurate spatiotemporal correlation.
• Implemented an efficient optimization algorithm using variable splitting and ADMM for model solving.

Main Results:

• MALLRT demonstrated promising performance on multiple cardiac MRI datasets (perfusion and cine) with retrospective and prospective undersampling.
• Achieved substantially better image reconstruction quality compared to four state-of-the-art methods.
• Quantified improvements using Signal to Error Ratio (SER) and Structural Similarity Index (SSIM) metrics, alongside visual perception.

Conclusions:

• The MALLRT model effectively addresses limitations of previous methods in dynamic cardiac MRI reconstruction.
• It significantly enhances image quality by preserving spatial details and capturing temporal variations.
• MALLRT offers a robust solution for reconstructing high-quality dynamic cardiac MRI from undersampled data.