Self-Learned Kernel Low Rank Approach TO Accelerated High Resolution 3D Diffusion MRI

TL;DR

This paper introduces a self-learned kernel low-rank method to accelerate high-resolution 3D diffusion MRI reconstruction from undersampled data, outperforming conventional compressed sensing techniques in quality and diffusion map accuracy.

Contribution

The paper presents a novel kernel low-rank approach that effectively reconstructs high-resolution 3D diffusion MRI from highly undersampled data, reducing acquisition time.

Findings

Outperforms conventional compressed sensing methods in image quality

Produces more accurate diffusion maps

Enables faster high-resolution 3D dMRI acquisition

Abstract

Diffusion Magnetic Resonance Imaging (dMRI) is a promising method to analyze the subtle changes in the tissue structure. However, the lengthy acquisition time is a major limitation in the clinical application of dMRI. Different image acquisition techniques such as parallel imaging, compressed sensing, has shortened the prolonged acquisition time but creating high-resolution 3D dMRI slices still requires a significant amount of time. In this study, we have shown that high-resolution 3D dMRI can be reconstructed from the highly undersampled k-space and q-space data using a Kernel LowRank method. Our proposed method has outperformed the conventional CS methods in terms of both image quality and diffusion maps constructed from the diffusion-weighted images