Acceleration of Multi-Direction Diffusion MRI Using EPI with Keyhole (EPIK) and Locally Low-Rank Tensor Reconstruction

Diffusion MRI is a commonly used non-invasive imaging technique with wide range of clinical applications. Many of these applications benefit from high spatial and angular resolution acquisitions as they increase the accuracy and precision of the generated quantitative maps. Higher resolution is attainable by using multi-shot EPI, which increases the scan time. To maintain the same scan efficiency as single-shot EPI, we propose an end-to-end framework that comprises an accelerated EPI-with-keyhole (EPIK) acquisition and a novel reconstruction algorithm based on space-coil-contrast locally low-rank tensor (SCC-LLRT) regularization. We compared the proposed SCC-LLRT algorithm with SPA-LLR for reconstructing EPIK data, and compared the proposed framework with a more traditional acceleration framework based on a combination of regular EPI and SPA-LLR, using retrospectively undersampled data from a healthy subject. The results showed that SCC-LLRT generated more accurate reconstruction with better preservation of high-frequency structural details compared with SPA-LLR. Furthermore, the proposed framework (EPIK+SCC-LLRT) achieved a substantially improved NRMSE, SSIM and reduction of ghosting artifacts compared with the traditional acceleration framework (EPI+SCC-LLRT). The results suggest that the proposed framework has a high feasibility for accelerating multi-direction diffusion MRI.

I. Introduction

Diffusion MRI (dMRI) is a non-invasive imaging technique that plays a pivotal role in both clinical practice and research within the fields of neuroscience and neurology^{[1] [2]}. Quantitative metrics derived from dMRI, such as mean diffusivity (MD) and fractional anisotropy (FA), hold immense diagnostic value in the detection of neural diseases such as acute ischemic stroke ^[3] and assessment of Alzheimer’s disease ^[4].