A Reconstruction Algorithm for Cherenkov‑Excited Luminescence Scanning Imaging Based on Unrolled Iterative Optimization

Mengfan Geng; Hu Zhang; Zhe Li; Ting Hu; Kebin Jia; Zhonghua Sun; Jinchao Feng

doi:10.3788/CJL230640

Chinese Journal of Lasers, Volume. 50, Issue 15, 1507106(2023)

A Reconstruction Algorithm for Cherenkov‑Excited Luminescence Scanning Imaging Based on Unrolled Iterative Optimization

Mengfan Geng^1,2, Hu Zhang^1,2, Zhe Li^1,2、**, Ting Hu^1,2, Kebin Jia^1,2, Zhonghua Sun^1,2, and Jinchao Feng^1,2、*

¹Beijing Key Laboratory of Computational Intelligence and Intelligent System, Faculty of Information Technology, Beijing University of Technology, Beijing 100124, China

²Beijing Laboratory of Advanced Information Networks, Beijing 100124, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(14)

Fig. 1. Schematic of the proposed ADMM-Net structure

Download full size

Fig. 2. Intermediate results of ADMM-Net iteration, (a)-(e) are the results of 1-5 iterations, respectively

Download full size

Fig. 3. Impact of filters on the reconstruction results. (a) Ground truth images; (b) BP reconstructed images; (c) FBP reconstructed images; (d) BP-ADMM reconstructed images; (e) ADMM-Net reconstructed images

Download full size

Fig. 4. Impact of symmetric loss on the reconstruction results. (a) Ground truth images; (b) ADMM-Net* reconstructed images; (c) ADMM-Net reconstructed images

Download full size

Fig. 5. Results of different algorithms for reconstructing a single fluorescent target with different contrasts. (a) Contrast is 4∶1;(b) contrast is 3.5∶1; (c) contrast is 3∶1; (d) contrast is 2.5∶1

Download full size

Fig. 6. Reconstruction results of two fluorescent targets with different algorithms. (a) Ground truth images; (b) sinogram signals; (c) FBP reconstructed images; (d) FBPConvNet reconstructed images; (e) ISTA-Net⁺ reconstructed images; (f) ADMM-Net reconstructed images

Download full size

Fig. 7. Statistic results of different algorithms reconstructed images. (a) RMSE; (b) PSNR; (c) SSIM

Download full size

Fig. 8. Different algorithms reconstructed two fluorescent targets with different edge-to-edge distances (the edge-to-edge distance of two targets from the top to the bottom rows is 1, 2, 4 and 6 mm, respectively). (a) Ground truth images; (b) FBP reconstructed images; (c) FBPConvNet reconstructed images; (d) ISTA-Net⁺ reconstructed images; (e) ADMM-Net reconstructed images

Download full size

Fig. 9. Fluorescence quantum yield profiles of two fluorescent targets with different edge-to-edge distances along the red dotted lines in Fig.8. (a) 1 mm; (b) 2 mm; (c) 4 mm; (d) 6 mm

Download full size

Fig. 10. Quantitative results of different algorithms reconstructed images. (a) RMSE; (b) PSNR; (c) SSIM

Download full size

Fig. 11. Reconstruction results with three and four fluorescent targets with different fluorescence quantum yield ratios for different algorithms (the quantum yield ratio of three targets is 4∶2∶1, while 7∶6∶5∶4 for four targets). (a) Ground truth images; (b) sinogram signals; (c) FBP reconstructed images; (d) FBPConvNet reconstructed images; (e) ISTA-Net⁺ reconstructed images; (f) ADMM-Net reconstructed images

Download full size

Table 1. Quantitative reconstruction results with different numbers of network layer
View table
Table 1. Quantitative reconstruction results with different numbers of network layer
Layer quantity RMSE PSNR/dB SSIM
3 1.51×10^-5 33.46 0.87
4 1.35×10^-5 34.46 0.88
5 1.24×10^-5 35.38 0.90
6 1.23×10^-5 36.05 0.90

Table 2. Quantitative reconstructed results for different algorithms under different contrasts

View table

Table 2. Quantitative reconstructed results for different algorithms under different contrasts

Contrast	Algorithm	RMSE	PSNR /dB	SSIM
4∶1	FBP	$7.55 \times 10^{- 5}$	20.50	0.61
	FBPConvNet	$1.42 \times 10^{- 5}$	34.99	0.84
	ISTA-Net⁺	$1.79 \times 10^{- 5}$	33.02	0.88
	ADMM-Net	$1.05 \times 10^{- 5}$	37.64	0.92
3.5∶1	FBP	$7.29 \times 10^{- 5}$	19.65	0.60
	FBPConvNet	$1.49 \times 10^{- 5}$	33.43	0.83
	ISTA-Net⁺	$1.81 \times 10^{- 5}$	31.73	0.85
	ADMM-Net	$1.20 \times 10^{- 5}$	35.34	0.87
3∶1	FBP	$7.07 \times 10^{- 5}$	18.58	0.58
	FBPConvNet	$1.62 \times 10^{- 5}$	31.37	0.83
	ISTA-Net⁺	$1.80 \times 10^{- 5}$	30.46	0.83
	ADMM-Net	$1.41 \times 10^{- 5}$	32.60	0.84
2.5∶1	FBP	$6.89 \times 10^{- 5}$	17.21	0.56
	FBPConvNet	$1.85 \times 10^{- 5}$	28.64	0.81
	ISTA-Net⁺	$1.74 \times 10^{- 5}$	29.16	0.81
	ADMM-Net	$1.68 \times 10^{- 5}$	29.45	0.83

Table 3. Quantitative results of reconstructed images with three and four fluorescent targets with different fluorescence quantum yield ratios for different algorithms

View table

Table 3. Quantitative results of reconstructed images with three and four fluorescent targets with different fluorescence quantum yield ratios for different algorithms

Ratio	Algorithm	RMSE	PSNR /dB	SSIM
4∶2∶1	FBP	$8.29 \times 10^{- 5}$	19.69	0.57
	FBPConvNet	$2.08 \times 10^{- 5}$	31.69	0.87
	ISTA-Net⁺	$2.45 \times 10^{- 5}$	30.28	0.91
	ADMM-Net	$1.97 \times 10^{- 5}$	32.18	0.94
7∶6∶5∶4	FBP	$9.85 \times 10^{- 5}$	18.19	0.52
	FBPConvNet	$4.12 \times 10^{- 5}$	25.77	0.89
	ISTA-Net⁺	$3.36 \times 10^{- 5}$	27.52	0.88
	ADMM-Net	$2.89 \times 10^{- 5}$	28.85	0.90

Tools

Get Citation

Copy Citation Text

Mengfan Geng, Hu Zhang, Zhe Li, Ting Hu, Kebin Jia, Zhonghua Sun, Jinchao Feng. A Reconstruction Algorithm for Cherenkov‑Excited Luminescence Scanning Imaging Based on Unrolled Iterative Optimization[J]. Chinese Journal of Lasers, 2023, 50(15): 1507106

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Biomedical Optical Imaging

Received: Mar. 22, 2023

Accepted: Apr. 25, 2023

Published Online: Aug. 8, 2023

The Author Email: Li Zhe (lizhe1023@bjut.edu.cn), Feng Jinchao (fengjc@bjut.edu.cn)

DOI:10.3788/CJL230640

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Quantitative reconstruction results with different numbers of network layer

Table 1. Quantitative reconstruction results with different numbers of network layer

Table 2. Quantitative reconstructed results for different algorithms under different contrasts

Table 2. Quantitative reconstructed results for different algorithms under different contrasts

Table 3. Quantitative results of reconstructed images with three and four fluorescent targets with different fluorescence quantum yield ratios for different algorithms

Table 3. Quantitative results of reconstructed images with three and four fluorescent targets with different fluorescence quantum yield ratios for different algorithms