Combinatorial Reconstruction and Segmentation of Magnetic Resonance Image Using Teacher Forcing

Yu Zhang; Haoran Li; Cheng Li; Fei Li; Shanshan Wang

doi:10.3788/LOP202259.1415024

Laser & Optoelectronics Progress, Volume. 59, Issue 14, 1415024(2022)

Combinatorial Reconstruction and Segmentation of Magnetic Resonance Image Using Teacher Forcing

Yu Zhang^1,2, Haoran Li^1,2, Cheng Li¹, Fei Li¹, and Shanshan Wang^1、*

¹Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen518055, Guangdong , China

²University of Chinese Academy of Sciences, Beijing 100049, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Framework of the proposed multi-task MRI method

Download full size

Fig. 2. The proposed improved teacher forcing scheme (TFS)

Download full size

Fig. 3. Examples of 1D random K-space mask with different acceleration factors. (a) Mask under 4× acceleration; (b) mask under 8× acceleration

Download full size

Fig. 4. Examples of lesion segmentation results and image reconstruction error maps of samples on the ATLAS dataset (acceleration factor is 4), from left to right is annotation, ours, SegNetMRI, U-Net (top) and D5C5 (bottom), SynNet, LI-Net, SERANet

Download full size

Fig. 5. Violin-plot of the segmentation results on the ATLAS dataset

Download full size

Fig. 6. Boxplot of the segmentation results on the ATLAS dataset

Download full size

Fig. 7. Examples of lesion segmentation results and image reconstruction error maps of samples on the in-house dataset (acceleration factor is 4), from left to right is label, ours, SegNetMRI, U-Net (top) and D5C5 (bottom)

Download full size

Table 1. Experimental results of different methods on the ATLAS dataset (acceleration factor is 4), bold letters indicate the best result

View table

Table 1. Experimental results of different methods on the ATLAS dataset (acceleration factor is 4), bold letters indicate the best result

Method	Dice	Precision	Recall	PSNR /dB	SSIM
SynNet	0.331±0.055	0.473±0.090	0.318±0.065
LI-Net	0.338±0.060	0.494±0.095	0.303±0.060
SERANet	0.353±0.072	0.682±0.126	0.280±0.060
SegNetMRI	0.224±0.067	0.618±0.164	0.156±0.034	28.22±0.746	0.950±0.00600
U-Net	0.451±0.070	0.641±0.085	0.409±0.075
D5C5				28.78±0.689	0.958±0.00005
Ours	0.515±0.067	0.665±0.079	0.490±0.077	28.88±0.763	0.959±0.00005

Table 2. Experimental results of different methods on the ATLAS dataset (acceleration factor is 8)

View table

Table 2. Experimental results of different methods on the ATLAS dataset (acceleration factor is 8)

Method	Dice	Precision	Recall	PSNR /dB	SSIM
SynNet	0.311±0.063	0.506±0.098	0.268±0.057
LI-Net	0.312±0.056	0.458±0.095	0.288±0.059
SERANet	0.303±0.080	0.582±0.151	0.245±0.065
SegNetMRI	0.081±0.019	0.368±0.197	0.050±0.009	22.96±0.678	0.887±0.0001
U-Net	0.420±0.069	0.571±0.093	0.393±0.075
D5C5				22.98±0.666	0.888±0.0002
Ours	0.445±0.073	0.628±0.095	0.407±0.077	23.04±0.610	0.892±0.0002

Table 3. Ablation study on the ATLAS dataset of effectiveness evaluation of the proposed teacher forcing scheme (acceleration factor is 4)
View table
Table 3. Ablation study on the ATLAS dataset of effectiveness evaluation of the proposed teacher forcing scheme (acceleration factor is 4)
TFS Dice Precision Recall PSNR /dB SSIM
0.494±0.071 0.636±0.135 0.462±0.067 28.78±0.732 0.957±0.00005
√ 0.515±0.067 0.665±0.079 0.490±0.077 28.88±0.763 0.959±0.00005

Table 4. Experimental results of different methods on the in-house dataset (acceleration factor is 4)

View table

Table 4. Experimental results of different methods on the in-house dataset (acceleration factor is 4)

Method	Dice	Precision	Recall	PSNR /dB	SSIM
SegNetMRI	0.841±0.004	0.814±0.024	0.859±0.033	32.52±0.057	0.983±0.006
U-Net	0.854±0.015	0.830±0.007	0.789±0.107
D5C5				32.49±0.044	0.984±0.008
Ours	0.864±0.001	0.880±0.011	0.868±0.031	32.65±0.013	0.986±0.001

Tools

Get Citation

Copy Citation Text

Yu Zhang, Haoran Li, Cheng Li, Fei Li, Shanshan Wang. Combinatorial Reconstruction and Segmentation of Magnetic Resonance Image Using Teacher Forcing[J]. Laser & Optoelectronics Progress, 2022, 59(14): 1415024

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Machine Vision

Received: Dec. 15, 2021

Accepted: Feb. 21, 2022

Published Online: Jul. 1, 2022

The Author Email: Shanshan Wang (ss.wang@siat.ac.cn)

DOI:10.3788/LOP202259.1415024

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Experimental results of different methods on the ATLAS dataset (acceleration factor is 4), bold letters indicate the best result

Table 1. Experimental results of different methods on the ATLAS dataset (acceleration factor is 4), bold letters indicate the best result

Table 2. Experimental results of different methods on the ATLAS dataset (acceleration factor is 8)

Table 2. Experimental results of different methods on the ATLAS dataset (acceleration factor is 8)

Table 3. Ablation study on the ATLAS dataset of effectiveness evaluation of the proposed teacher forcing scheme (acceleration factor is 4)

Table 3. Ablation study on the ATLAS dataset of effectiveness evaluation of the proposed teacher forcing scheme (acceleration factor is 4)

Table 4. Experimental results of different methods on the in-house dataset (acceleration factor is 4)

Table 4. Experimental results of different methods on the in-house dataset (acceleration factor is 4)