Model-Based Deep Learning for Accelerated Diffusion MRI
Deep learning has been applied to accelerate diffusion MRI acquisitions in the past. While the prospect of fast multi-shell dMRI is exciting, there remains concerns about the generalizability of DL approaches to translate to specific acquisition demands. We present a new approach to mitigate such concerns. We combine the power of traditional model-based reconstructions and its flexibility to adapt to any acquisition settings, with the power of machine learning to utilize pre-learned information, to derive an efficient highly adaptable reconstruction framework for accelerated dMRI acquisition. Further, we use the learned information in a plug-and-play manner which provides even more flexibility. The proposed method is shown to have excellent performance in various acquisition settings and across field strengths.
 |
Discussion points of the webinar:
|
Merry Mani, PhD, Assistant Professor in Radiology
University of Iowa Carver College of Medicine
My work is focused on the development of advanced diffusion MRI acquisition methods. Works in the past include: the development of an efficient joint k-q accelerated diffusion MRI acquisition which provides a means to accelerate in multiple dimensions of acquisition as the same time. The above acquisition was formulated in the context of compressed sensing. We have also proposed joint k-q reconstruction methods that maximally exploits the SNR available from all the k-q measurements points. Other works include the development of a phase-artifact free multi-shot diffusion MRI reconstruction method. The method, dubbed MUSSELS, allows to recover high resolution diffusion data from multi-shot acquisitions without the need for navigators or explicit phase compensation. The formulation was based on a constrained parallel imaging, requiring only coil sensitivity information to obtain the phase-artifact free high resolution images. Recently, we have worked on utilizing the multi-shot sequences for multi-shell imaging, which utilizes joint k-q acceleration and employing model-based deep learning reconstruction derive high resolution, high SNR multi-shell data, without needing navigators. Merry Mani obtained her PhD in 2014 in Electrical Engineering from the University of Rochester in New York. In 2014, she joined the Magnetic Resonance Research Facility at the University of Iowa as a post-doctoral fellow working on the 7T MRI for diffusion multi-shot imaging. In 2019, she become an Assistant Professor in Radiology at the University of Iowa Carver College of Medicine.
