Improved MRF Reconstruction via Structure-Preserved Graph Embedding Framework.

Area of Science:

• Medical Imaging
• Computational Physics
• Data Science

Background:

• Undersampled magnetic resonance fingerprinting (MRF) schemes cause aliasing artifacts, reducing quantitative imaging accuracy.
• Current reconstruction methods often fail to leverage MRF data structure and exhibit high computational complexity.
• Pixel-wise data priors struggle with non-local and non-linear correlations.

Purpose of the Study:

• To develop a novel MRF reconstruction framework that exploits non-linear and non-local redundancies in MRF data.
• To address limitations of existing methods by reducing computational complexity and improving reconstruction quality.
• To enhance quantitative accuracy in MRF imaging.

Main Methods:

• A graph embedding framework is proposed, remodeling MRF data and parameter maps as graph nodes.
• The reconstruction problem is redefined as a structure-preserved graph embedding problem.
• A novel scheme for estimating graph structure is introduced, revealing low-dimensional representations of MRF data nodes.

Main Results:

• The framework preserves intrinsic graph structures between MRF data and parameter nodes, exploiting global graph properties.
• MRF data recovery and parameter map estimation are integrated into a single optimization problem.
• Substantial reduction in computational complexity is achieved, with minimal increase relative to data acquisition length.

Conclusions:

• The proposed graph embedding approach enables high-quality MRF data and parameter map reconstruction.
• The method significantly reduces computational time compared to existing techniques.
• This framework offers a promising solution for accurate and efficient quantitative MRF imaging.