yqx7150 / UGM-PI

Universal Generative Modeling for Calibration-free Parallel MR Imaging

UGM-PI

Paper: Universal Generative Modeling for Calibration-free Parallel MR Imaging

Authors: Wanqing Zhu, Bing Guan, Shanshan Wang, Minghui Zhang and Qiegen Liu

      Department of Electronic Information Engineering, Nanchang University      
      Paul C. Lauterbur Research Center for Biomedical Imaging, SIAT, CAS     

Paper #160 accepted at IEEE ISBI 2022.

Date : Feb-12-2022
Version : 1.0
The code and the algorithm are for non-comercial use only.
Copyright 2022, Department of Electronic Information Engineering, Nanchang University.

The integration of compressed sensing and parallel imaging (CS-PI) provides a robust mechanism for accelerating MRI acquisitions. However, most such strategies require the ex-plicit formation of either coil sensitivity profiles or a cross-coil correlation operator, and as a result reconstruction corresponds to solving a challenging bilinear optimization problem. In this work, we present an unsupervised deep learning framework for calibration-free parallel MRI, coined universal generative modeling for parallel imaging (UGM-PI). More precisely, we make use of the merits of both wavelet transform and the adaptive iteration strategy in a unified framework. We train a powerful noise condi-tional score network by forming wavelet tensor as the net-work input at the training phase. Experimental results on both physical phantom and in vivo datasets implied that the proposed method is comparable and even superior to state-of-the-art CS-PI reconstruction approaches.

Flowchart illustration of UGM-PI

Top: Network training process in single coil image for prior learning. Bottom: An intermediate result of the iterative process at multi-coil MRI reconstruction phase.

Convergence tendency comparison

Convergence tendency comparison of DSM in the native NCSN and the advanced UGM-PI, respectively.

Reconstruction comparison 1

The test comparison of 4 coils physical phantom obtained by P-LORAKS, SAKE, LINDBREG, and the proposed UGM-PI under 2D random sampling with the acceleration factor R=6.

Reconstruction comparison 2

Reconstruction comparison of the in vivo MPRAGE brain image obtained by P-LORAKS, SAKE, LINDBREG, and UGM-PI under 2D variable density Poisson sampling with the acceleration factor R=10.

Training

python3 separate_SIAT.py --model ugm_pi --runner SIAT_Train --config iOrth_8h_10sigma.yml --doc iOrth_8h_10sigma

Test

python3 separate_SIAT.py --model ugm_pi --runner SIAT_Train --doc iOrth_8h_10sigma --test

Checkpoints

We provide pretrained checkpoints. You can download pretrained models from Baidu Drive. key number is "7150"

Other Related Projects

• Homotopic Gradients of Generative Density Priors for MR Image Reconstruction
  [Paper] [Code] [PPT]

• Iterative Reconstruction for Low-Dose CT using Deep Gradient Priors of Generative Model
  [Paper] [Code] [PPT]

• Progressive Colorization via Interative Generative Models
  [Paper] [Code] [PPT]

• Joint Intensity-Gradient Guided Generative Modeling for Colorization [Paper] [Code] [PPT] [数学图像联盟会议交流PPT]

• Wavelet Transform-assisted Adaptive Generative Modeling for Colorization [Paper] [Code] [数学图像联盟会议交流PPT]