Reconstruction of Magnetic Resonance Imaging Based on Dual-Domain Densely-Connected Residual Convolutional Networks

Weikun Zhang; Qiaohong Liu; Xiaoxiang Han; Yuanjie Lin; Keyan Chen

doi:10.3788/LOP231468

Laser & Optoelectronics Progress, Volume. 61, Issue 12, 1211005(2024)

Reconstruction of Magnetic Resonance Imaging Based on Dual-Domain Densely-Connected Residual Convolutional Networks

Weikun Zhang¹, Qiaohong Liu^2、*, Xiaoxiang Han¹, Yuanjie Lin¹, and Keyan Chen¹

Author Affiliations

¹School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China

²College of Medical Instruments, Shanghai University of Medicine & Health Sciences, Shanghai 201318, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Algorithm flow chart. (a) Fully-sampled K-space; (b) under-sampled K-space; (c) reconstruction of the K-space; (d) initial reconstruction of the image; (e) reconstructed image

Download full size

Fig. 2. Overall architecture of DDCRNet

Download full size

Fig. 3. Structure of densely-connected residual block

Download full size

Fig. 4. Modular structure of coordinate attention

Download full size

Fig. 5. Examples of three under-sampled masks. (a) 2D Gaussian mask; (b) 1D Gaussian mask; (c) radial mask

Download full size

Fig. 6. Qualitative results of a two-domain ablation experiment. (a) Fully-sampled image and under-sampled image under the 2D Gaussian mask with sampling rate of 10%; (b) reconstruction and error diagrams of K-Net; (c) reconstruction and error diagrams of I-Net; (d) reconstruction and error diagrams of DDCRNet

Download full size

Fig. 7. Reconstructed qualitative results of different algorithms. (a) Fully-sampled image and under-sampled image; (b) UNet; (c) KIKINet; (d) XPDNet; (e) WNet; (f) MDReconNet; (g) DDCRNet

Download full size

Fig. 8. Reconstructed qualitative results of different algorithms. (a) Fully-sampled image and under-sampled image; (b) UNet; (c) KIKINet; (d) XPDNet; (e) WNet; (f) MDReconNet; (g) DDCRNet

Download full size

Table 1. Quantitative results of a two-domain ablation experiment under the two-dimensional Gaussian mask with sampling rate of 10%
View table
Table 1. Quantitative results of a two-domain ablation experiment under the two-dimensional Gaussian mask with sampling rate of 10%
Method SSIM NRMSE /% PSNR /dB
K-Net 0.7848 5.740 24.83
I-Net 0.8585 5.456 25.52
DDCRNet 0.9158 3.794 28.69

Table 2. Comparison of 2D Gaussian mask reconstruction results with different sampling rates

View table

Table 2. Comparison of 2D Gaussian mask reconstruction results with different sampling rates

Network	10% sampling rate			20% sampling rate			30% sampling rate
Network	SSIM	NRMSE /%	PSNR /dB	SSIM	NRMSE /%	PSNR /dB	SSIM	NRMSE /%	PSNR /dB
Zero filled	0.6048	11.43	18.85	0.6821	9.426	20.54	0.7582	7.487	22.54
UNet	0.6654	6.513	23.88	0.8606	3.965	28.07	0.9110	3.085	30.24
KIKINet	0.8342	6.577	23.88	0.8543	6.597	24.20	0.8748	6.081	24.90
XPDNet	0.8585	5.434	25.44	0.8918	4.008	28.01	0.9169	3.176	30.00
WNet	0.8790	5.108	26.14	0.9306	3.051	30.48	0.9485	2.266	33.05
MDReconNet	0.8616	4.611	27.02	0.8971	3.233	30.11	0.9466	2.642	31.83
DDCRNet	0.9158	3.794	28.69	0.9373	2.917	30.85	0.9526	2.200	33.25

Table 3. Reconstructed quantitative results of different algorithms under the different masks

View table

Table 3. Reconstructed quantitative results of different algorithms under the different masks

Network	1D Gaussian mask			Radial mask
Network	SSIM	NRMSE /%	PSNR /dB	SSIM	NRMSE /%	PSNR /dB
Zero filled	0.8117	7.444	22.57	0.7504	6.709	23.51
UNet	0.9064	3.626	28.87	0.9063	3.229	29.90
KIKINet	0.8732	6.766	23.94	0.8602	7.825	22.79
XPDNet	0.9154	4.180	27.68	0.9160	4.044	28.03
WNet	0.9347	2.769	31.23	0.9339	3.342	29.76
MDReconNet	0.9290	2.750	31.41	0.9335	3.125	30.49
DDCRNet	0.9486	2.681	31.55	0.9417	3.077	30.42

Tools

Get Citation

Copy Citation Text

Weikun Zhang, Qiaohong Liu, Xiaoxiang Han, Yuanjie Lin, Keyan Chen. Reconstruction of Magnetic Resonance Imaging Based on Dual-Domain Densely-Connected Residual Convolutional Networks[J]. Laser & Optoelectronics Progress, 2024, 61(12): 1211005

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Imaging Systems

Received: Jun. 6, 2023

Accepted: Aug. 22, 2023

Published Online: Jun. 3, 2024

The Author Email: Qiaohong Liu (hqllqh@163.com)

DOI:10.3788/LOP231468

CSTR:32186.14.LOP231468

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Quantitative results of a two-domain ablation experiment under the two-dimensional Gaussian mask with sampling rate of 10%

Table 1. Quantitative results of a two-domain ablation experiment under the two-dimensional Gaussian mask with sampling rate of 10%

Table 2. Comparison of 2D Gaussian mask reconstruction results with different sampling rates

Table 2. Comparison of 2D Gaussian mask reconstruction results with different sampling rates

Table 3. Reconstructed quantitative results of different algorithms under the different masks

Table 3. Reconstructed quantitative results of different algorithms under the different masks