Undersampled MR image reconstruction using an enhanced recursive residual network

Lijun Bao, Fuze Ye, Congbo Cai, Jian Wu, Kun Zeng, Peter C Van Zijl, Zhong Chen

Research output: Contribution to journalArticle

Abstract

When using aggressive undersampling, it is difficult to recover the high quality image with reliably fine features. In this paper, we propose an enhanced recursive residual network (ERRN) that improves the basic recursive residual network with a high-frequency feature guidance, an error-correction unit and dense connections. The feature guidance is designed to predict the underlying anatomy based on image a priori learned from the label data, playing a complementary role to the residual learning. The ERRN is adapted for two important applications: compressed sensing (CS) MRI and super resolution (SR) MRI, while an application-specific error-correction unit is added into the framework, i.e. data consistency for CS-MRI and back projection for SR-MRI due to their different sampling schemes. Our proposed network was evaluated using a real-valued brain dataset, a complex-valued knee dataset, pathological brain data and in vivo rat brain data with different undersampling masks and rates. Experimental results demonstrated that ERRN presented superior reconstructions at all cases with distinctly restored structural features and highest image quality metrics compared to both the state-of-the-art convolutional neural networks and the conventional optimization-based methods, particularly for the undersampling rate over 5-fold. Thus, an excellent framework design can endow the network with a flexible architecture, fewer parameters, outstanding performances for various undersampling schemes, and reduced overfitting in generalization, which will facilitate real-time reconstruction on MRI scanners.

Original languageEnglish (US)
Pages (from-to)232-246
Number of pages15
JournalJournal of Magnetic Resonance
Volume305
DOIs
StatePublished - Aug 1 2019

Fingerprint

Computer-Assisted Image Processing
image reconstruction
Image reconstruction
Magnetic resonance imaging
Brain
Compressed sensing
Error correction
brain
Image quality
Masks
Knee
Anatomy
Learning
Rats
Labels
anatomy
Sampling
Neural networks
learning
scanners

Keywords

  • Convolutional neural network
  • Error-correction
  • Feature guidance
  • Recursive residual learning
  • Undersampled MRI reconstruction

ASJC Scopus subject areas

  • Biophysics
  • Biochemistry
  • Nuclear and High Energy Physics
  • Condensed Matter Physics

Cite this

Undersampled MR image reconstruction using an enhanced recursive residual network. / Bao, Lijun; Ye, Fuze; Cai, Congbo; Wu, Jian; Zeng, Kun; Van Zijl, Peter C; Chen, Zhong.

In: Journal of Magnetic Resonance, Vol. 305, 01.08.2019, p. 232-246.

Research output: Contribution to journalArticle

Bao, Lijun ; Ye, Fuze ; Cai, Congbo ; Wu, Jian ; Zeng, Kun ; Van Zijl, Peter C ; Chen, Zhong. / Undersampled MR image reconstruction using an enhanced recursive residual network. In: Journal of Magnetic Resonance. 2019 ; Vol. 305. pp. 232-246.
@article{a5cf75777a1547e494dd3506412aa042,
title = "Undersampled MR image reconstruction using an enhanced recursive residual network",
abstract = "When using aggressive undersampling, it is difficult to recover the high quality image with reliably fine features. In this paper, we propose an enhanced recursive residual network (ERRN) that improves the basic recursive residual network with a high-frequency feature guidance, an error-correction unit and dense connections. The feature guidance is designed to predict the underlying anatomy based on image a priori learned from the label data, playing a complementary role to the residual learning. The ERRN is adapted for two important applications: compressed sensing (CS) MRI and super resolution (SR) MRI, while an application-specific error-correction unit is added into the framework, i.e. data consistency for CS-MRI and back projection for SR-MRI due to their different sampling schemes. Our proposed network was evaluated using a real-valued brain dataset, a complex-valued knee dataset, pathological brain data and in vivo rat brain data with different undersampling masks and rates. Experimental results demonstrated that ERRN presented superior reconstructions at all cases with distinctly restored structural features and highest image quality metrics compared to both the state-of-the-art convolutional neural networks and the conventional optimization-based methods, particularly for the undersampling rate over 5-fold. Thus, an excellent framework design can endow the network with a flexible architecture, fewer parameters, outstanding performances for various undersampling schemes, and reduced overfitting in generalization, which will facilitate real-time reconstruction on MRI scanners.",
keywords = "Convolutional neural network, Error-correction, Feature guidance, Recursive residual learning, Undersampled MRI reconstruction",
author = "Lijun Bao and Fuze Ye and Congbo Cai and Jian Wu and Kun Zeng and {Van Zijl}, {Peter C} and Zhong Chen",
year = "2019",
month = "8",
day = "1",
doi = "10.1016/j.jmr.2019.07.020",
language = "English (US)",
volume = "305",
pages = "232--246",
journal = "Journal of Magnetic Resonance",
issn = "1090-7807",
publisher = "Academic Press Inc.",

}

TY - JOUR

T1 - Undersampled MR image reconstruction using an enhanced recursive residual network

AU - Bao, Lijun

AU - Ye, Fuze

AU - Cai, Congbo

AU - Wu, Jian

AU - Zeng, Kun

AU - Van Zijl, Peter C

AU - Chen, Zhong

PY - 2019/8/1

Y1 - 2019/8/1

N2 - When using aggressive undersampling, it is difficult to recover the high quality image with reliably fine features. In this paper, we propose an enhanced recursive residual network (ERRN) that improves the basic recursive residual network with a high-frequency feature guidance, an error-correction unit and dense connections. The feature guidance is designed to predict the underlying anatomy based on image a priori learned from the label data, playing a complementary role to the residual learning. The ERRN is adapted for two important applications: compressed sensing (CS) MRI and super resolution (SR) MRI, while an application-specific error-correction unit is added into the framework, i.e. data consistency for CS-MRI and back projection for SR-MRI due to their different sampling schemes. Our proposed network was evaluated using a real-valued brain dataset, a complex-valued knee dataset, pathological brain data and in vivo rat brain data with different undersampling masks and rates. Experimental results demonstrated that ERRN presented superior reconstructions at all cases with distinctly restored structural features and highest image quality metrics compared to both the state-of-the-art convolutional neural networks and the conventional optimization-based methods, particularly for the undersampling rate over 5-fold. Thus, an excellent framework design can endow the network with a flexible architecture, fewer parameters, outstanding performances for various undersampling schemes, and reduced overfitting in generalization, which will facilitate real-time reconstruction on MRI scanners.

AB - When using aggressive undersampling, it is difficult to recover the high quality image with reliably fine features. In this paper, we propose an enhanced recursive residual network (ERRN) that improves the basic recursive residual network with a high-frequency feature guidance, an error-correction unit and dense connections. The feature guidance is designed to predict the underlying anatomy based on image a priori learned from the label data, playing a complementary role to the residual learning. The ERRN is adapted for two important applications: compressed sensing (CS) MRI and super resolution (SR) MRI, while an application-specific error-correction unit is added into the framework, i.e. data consistency for CS-MRI and back projection for SR-MRI due to their different sampling schemes. Our proposed network was evaluated using a real-valued brain dataset, a complex-valued knee dataset, pathological brain data and in vivo rat brain data with different undersampling masks and rates. Experimental results demonstrated that ERRN presented superior reconstructions at all cases with distinctly restored structural features and highest image quality metrics compared to both the state-of-the-art convolutional neural networks and the conventional optimization-based methods, particularly for the undersampling rate over 5-fold. Thus, an excellent framework design can endow the network with a flexible architecture, fewer parameters, outstanding performances for various undersampling schemes, and reduced overfitting in generalization, which will facilitate real-time reconstruction on MRI scanners.

KW - Convolutional neural network

KW - Error-correction

KW - Feature guidance

KW - Recursive residual learning

KW - Undersampled MRI reconstruction

UR - http://www.scopus.com/inward/record.url?scp=85068936654&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85068936654&partnerID=8YFLogxK

U2 - 10.1016/j.jmr.2019.07.020

DO - 10.1016/j.jmr.2019.07.020

M3 - Article

C2 - 31323504

AN - SCOPUS:85068936654

VL - 305

SP - 232

EP - 246

JO - Journal of Magnetic Resonance

JF - Journal of Magnetic Resonance

SN - 1090-7807

ER -