RUN-UP: Accelerated Multi-Shot DWI Reconstruction Using an Unrolled Network with U-Net as Priors

In this work, we accelerate and improve the reconstruction of multi-shot diffusion-weighted MRI by an unrolled pipeline, in which the presumed regularization term is replaced by a U-Net. The proposed pipeline contains recurrences of model-based gradient updates and neural networks alternating between image space and data acquisition space. Several new features (using results of the joint reconstruction method as the learning target, including non-diffusion-weighted images as inputs, using Parametric ReLU and intermediate loss) are also introduced, and improvements of those features are demonstrated in an ablation study. Almost real-time reconstruction is achieved with the denoising effect similar to averaging of multiple repetitions. The generalizability of the proposed method is also demonstrated in the breast data.

Hu Y, Xu Y, Tian Q, Chen F, Shi X, Moran CJ, Daniel BL, Hargreaves BA. RUN-UP: Accelerated multishot diffusion-weighted MRI reconstruction using an unrolled network with U-Net as priors. Magn Reson Med. 2021 Feb;85(2):709-720.

Online Journal Article

Comparison of images of multi-shot DWI data reconstructed using different techniques: Shot-LLR, RUN-UP, and joint reconstruction.

Senior Research Scientist - Physical, Rad/Radiological Sciences Laboratory
Professor of Radiology (Body Imaging) and, by courtesy, of Bioengineering
Professor of Radiology (Radiological Sciences Laboratory) and, by courtesy, of Electrical Engineering and of Bioengineering
This profile is not available