Structure Preserving Compressive Sensing MRI Reconstruction using Generative Adversarial Networks

Puneesh Deora*^   Bhavya Vasudeva*^   Saumik Bhattacharya**   P. M. Pradhan*  
*Indian Institute of Technology Roorkee
**Indian Institute of Technology Kharagpur
The IEEE/CVF CVPR Workshop on New Trends in Image Restoration and Enhancement (NTIRE) 2020
^ equal contribution

Abstract

Compressive sensing magnetic resonance imaging (CS-MRI) accelerates the acquisition of MR images by breaking the Nyquist sampling limit. In this work, a novel generative adversarial network (GAN) based framework for CS-MRI reconstruction is proposed. Leveraging a combination of patch-based discriminator and structural similarity index based loss, our model focuses on preserving high frequency content as well as fine textural details in the reconstructed image. Dense and residual connections have been incorporated in a U-net based generator architecture to allow easier transfer of information as well as variable network length. We show that our algorithm outperforms state-of-the-art methods in terms of quality of reconstruction and robustness to noise. Also, the reconstruction time, which is of the order of milliseconds, makes it highly suitable for real-time clinical use.

Spotlight Video

Click the image to watch a 1 minute summary of our work.

IMAGE ALT TEXT

Network Architecture

(a) Generator and (b) discriminator architecture.

Results

Reconstruction results for 20% undersampled images. (a) GT, reconstruction results for (b) noise-free image, (c) image with 10% noise, and (d) image with 20% noise.

For more results, please refer the paper.

Paper

The paper can be found here.

Code

The official GitHub repository of this paper can be found here.

Citation

If you find our research useful, please cite our work.

@article{deora2019structure,
    title={Structure Preserving Compressive Sensing MRI Reconstruction using Generative Adversarial Networks},
    author={P. Deora and B. Vasudeva and S. Bhattacharya and P. M. Pradhan},
    journal={ArXiv},
    year={2019},
    volume={abs/1910.06067}
}