Self‐supervised learning of physics‐guided reconstruction neural networks without fully sampled reference data

Burhaneddin Yaman; Seyed Amir Hossein Hosseini; Steen Moeller; Jutta Ellermann; Kâmil Uğurbil; Mehmet Akçakaya

doi:10.1002/mrm.28378

journal article Open Access Jul 02, 2020

Self‐supervised learning of physics‐guided reconstruction neural networks without fully sampled reference data

Burhaneddin Yaman

Seyed Amir Hossein Hosseini

Steen Moeller

Jutta Ellermann Kâmil Uğurbil

Mehmet Akçakaya

Magnetic Resonance in Medicine Vol. 84 No. 6 pp. 3172-3191 · Wiley

View at Publisher Save 10.1002/mrm.28378

Abstract

PurposeTo develop a strategy for training a physics‐guided MRI reconstruction neural network without a database of fully sampled data sets.MethodsSelf‐supervised learning via data undersampling (SSDU) for physics‐guided deep learning reconstruction partitions available measurements into two disjoint sets, one of which is used in the data consistency (DC) units in the unrolled network and the other is used to define the loss for training. The proposed training without fully sampled data is compared with fully supervised training with ground‐truth data, as well as conventional compressed‐sensing and parallel imaging methods using the publicly available fastMRI knee database. The same physics‐guided neural network is used for both proposed SSDU and supervised training. The SSDU training is also applied to prospectively two‐fold accelerated high‐resolution brain data sets at different acceleration rates, and compared with parallel imaging.ResultsResults on five different knee sequences at an acceleration rate of 4 shows that the proposed self‐supervised approach performs closely with supervised learning, while significantly outperforming conventional compressed‐sensing and parallel imaging, as characterized by quantitative metrics and a clinical reader study. The results on prospectively subsampled brain data sets, in which supervised learning cannot be used due to lack of ground‐truth reference, show that the proposed self‐supervised approach successfully performs reconstruction at high acceleration rates (4, 6, and 8). Image readings indicate improved visual reconstruction quality with the proposed approach compared with parallel imaging at acquisition acceleration.ConclusionThe proposed SSDU approach allows training of physics‐guided deep learning MRI reconstruction without fully sampled data, while achieving comparable results with supervised deep learning MRI trained on fully sampled data.

Topics

No keywords indexed for this article. Browse by subject →

References

81

[1]

10.1002/mrm.10171

[2]

10.1002/(sici)1522-2594(199911)42:5<952::aid-mrm16>3.0.co;2-s

[3]

SPIRiT: Iterative self‐consistent parallel imaging reconstruction from arbitrary k‐space

Michael Lustig, John M. Pauly

Magnetic Resonance in Medicine 10.1002/mrm.22428

[4]

10.1002/mrm.21391

[5]

10.1109/tmi.2010.2085084

[6]

10.1109/tmi.2008.927346

[7]

10.1002/mrm.21202

[8]

10.1109/tmi.2010.2089519

[9]

10.1002/mrm.22841

[10]

10.1002/mrm.21236

[11]

10.1002/mrm.22161

[12]

10.1002/mrm.22463

[13]

10.1002/mrm.21728

[14]

10.1002/mrm.23279

[15]

10.1002/mrm.20133

[16]

10.1002/mrm.24592

[17]

10.1109/tpami.2018.2883941

[18]

10.1109/tmi.2018.2885599

[19]

10.1109/tip.2012.2195015

[20]

10.1109/msp.2019.2950557

[21]

WangS SuZ YingL et al.Accelerating magnetic resonance imaging via deep learning. In: Proceedings of the IEEE 13th International Symposium on Biomedical Imaging (ISBI) Prague Czech Republic 2016. pp514‐517. 10.1109/isbi.2016.7493320

[22]

10.1109/tbme.2018.2821699

[23]

10.1002/mrm.27420

[24]

10.1038/nature25988

[25]

10.1109/tmi.2018.2858752

[26]

10.1109/tmi.2019.2927101

[27]

10.1002/mrm.26977

[28]

MoDL: Model-Based Deep Learning Architecture for Inverse Problems

Hemant K. Aggarwal, Merry P. Mani, Mathews Jacob

IEEE Transactions on Medical Imaging 10.1109/tmi.2018.2865356

[29]

ZhangJ GhanemB.ISTA‐Net: Interpretable optimization‐inspired deep network for image compressive sensing. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Salt Lake City Utah 2018. pp1828‐1837. 10.1109/cvpr.2018.00196

[30]

Yang Y "Deep ADMM‐Net for compressive sensing MRI" Adv Neural Inf Process Syst (2016)

[31]

10.1109/tmi.2017.2760978

[32]

10.1109/tmi.2018.2863670

[33]

Mardani M "Neural proximal gradient descent for compressive imaging" Adv Neural Inf Process Syst (2018)

[34]

GregorK LeCunY.Learning fast approximations of sparse coding. In: Proceedings of the 27th International Conference on Machine Learning Haifa Israel 2010. pp399‐406.

[35]

Hosseini SAH "Dense recurrent neural networks for accelerated MRI: History‐cognizant unrolling of optimization algorithms" IEEE J Selec Top Sign Proc

[36]

ChengJY PaulyJM VasanawalaSS.Multi‐channel image reconstruction with latent coils and adverserial loss. In: Proceedings of the 27th Annual Meeting of ISMRM Montréal Canada 2019. Abstract 1249.

[37]

WangP ChenEZ ChenT PatelVM SunS.Pyramid convolutional RNN for MRI reconstruction. arXiv:1912.00543;2019.

[38]

10.1002/mrm.26918

[39]

10.1148/radiol.12110918

[40]

10.1186/1532-429x-16-2

[41]

10.1016/j.neuroimage.2013.05.012

[42]

10.1016/j.neuroimage.2013.05.078

[43]

k‐t FOCUSS: A general compressed sensing framework for high resolution dynamic MRI

Hong Jung, Kyunghyun Sung, Krishna S. Nayak et al.

Magnetic Resonance in Medicine 10.1002/mrm.21757

[44]

10.1002/mrm.21477

[45]

10.1002/mrm.22595

[46]

10.1109/tip.2012.2183143

[47]

10.1002/mrm.22483

[48]

10.1109/tmi.2010.2090538

[49]

10.1109/msp.2019.2943645

[50]

10.1109/tip.2010.2047910

Showing 50 of 81 references

Cited By

253

Masked cross-domain self-supervised deep learning framework for photoacoustic computed tomography reconstruction

Hengrong Lan, Lijie Huang · 2024

Neural Networks

PARCEL: Physics-Based Unsupervised Contrastive Representation Learning for Multi-Coil MR Imaging

Shanshan Wang, Ruoyou Wu · 2023

IEEE/ACM Transactions on Computatio...

Dual-domain self-supervised learning for accelerated non-Cartesian MRI reconstruction

Bo Zhou, Jo Schlemper · 2022

Medical Image Analysis

Metrics

253

Citations

81

References

Details

Published: Jul 02, 2020
Vol/Issue: 84(6)
Pages: 3172-3191
License: View

Authors

B

Burhaneddin Yaman

Department of Electrical and Computer Engineering University of Minnesota Minneapolis Minnesota USA; Center for Magnetic Resonance Research University of Minnesota Minneapolis Minnesota USA

S

Seyed Amir Hossein Hosseini

Department of Electrical and Computer Engineering University of Minnesota Minneapolis Minnesota USA; Center for Magnetic Resonance Research University of Minnesota Minneapolis Minnesota USA

S

Steen Moeller

Center for Magnetic Resonance Research University of Minnesota Minneapolis Minnesota USA

J

Jutta Ellermann

Center for Magnetic Resonance Research University of Minnesota Minneapolis Minnesota USA

K

Kâmil Uğurbil

Center for Magnetic Resonance Research University of Minnesota Minneapolis Minnesota USA

M

Mehmet Akçakaya

Department of Electrical and Computer Engineering University of Minnesota Minneapolis Minnesota USA; Center for Magnetic Resonance Research University of Minnesota Minneapolis Minnesota USA

Funding

Division of Computing and Communication Foundations Award: CAREER CCF‐1651825

National Institute of Biomedical Imaging and Bioengineering Award: P41EB027061

Cite This Article

Burhaneddin Yaman, Seyed Amir Hossein Hosseini, Steen Moeller, et al. (2020). Self‐supervised learning of physics‐guided reconstruction neural networks without fully sampled reference data. Magnetic Resonance in Medicine, 84(6), 3172-3191. https://doi.org/10.1002/mrm.28378

Self‐supervised learning of physics‐guided reconstruction neural networks without fully sampled reference data

You May Also Like