MLS: Joint Manifold-Learning and Sparsity-Aware Framework for Highly Accelerated Dynamic Magnetic Resonance Imaging

Area of Science:

• Medical Imaging
• Computer Vision
• Signal Processing

Background:

• Manifold-based models accelerate dynamic magnetic resonance imaging (dMRI).
• Conventional low-rank approaches are efficient, but joint low-rank and sparsity-aware modeling shows superior performance due to dMR image sparsity.

Purpose of the Study:

• To propose a novel joint manifold-learning and sparsity-aware framework for accelerating dMRI.
• To link recent manifold models with conventional sparsity-aware models for improved dMRI reconstruction.

Main Methods:

• Dynamic MR images are modeled as points on a smooth manifold.
• A data-driven manifold-learning approach preserves affine relations to learn low-dimensional embeddings.
• Temporal basis learning captures image periodicity, enforcing sparsity during reconstruction.

Main Results:

• The proposed framework effectively integrates manifold learning and sparsity.
• The learned temporal basis captures inherent periodicity in dynamic MR images.
• Validation on phantom and in-vivo data demonstrates the framework's efficacy.

Conclusions:

• The joint manifold-learning and sparsity-aware framework offers an efficient approach for dMRI acceleration.
• This method enhances reconstruction by leveraging both manifold geometry and temporal sparsity.
• The framework shows promise for improving the speed and quality of dynamic MRI acquisition.