Projection Domain Denoising Method for Multi-Energy Computed Tomography via Dual-Stream Transformer

Shunxin Ouyang; Zaifeng Shi; Fanning Kong; Lili Zhang; Qingjie Cao

doi:10.3788/LOP231426

Laser & Optoelectronics Progress, Volume. 61, Issue 8, 0837008(2024)

Projection Domain Denoising Method for Multi-Energy Computed Tomography via Dual-Stream Transformer

Shunxin Ouyang¹, Zaifeng Shi^1,3、*, Fanning Kong¹, Lili Zhang¹, and Qingjie Cao²

Author Affiliations

¹School of Microelectronics, Tianjin University, Tianjin 300072, China

²College of Mathematical Science, Tianjin Normal University, Tianjin 300387, China

³Tianjin Key Laboratory of Imaging and Sensing Microelectronics Technology, Tianjin 300072, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(11)

Fig. 1. Structure diagram of the proposed DTE-PDN-based low-dose multi-energy CT denoising method

Download full size

Fig. 2. Structure diagram of the dual-stream Transformer

Download full size

Fig. 3. Loss convergence curves on simulation dataset

Download full size

Fig. 4. Comparison of denoising effects of abdomen slice from different methods

Download full size

Fig. 5. Comparison of denoising effects of chest slice from different methods

Download full size

Fig. 6. Ablation experiment results of different loss function combinations. (a) Selected projection path; (b) linear attenuation coefficient curves

Download full size

Fig. 7. Curves of the evaluation index of the chest slice processed by DTE-PDN trained by the hybrid loss with different weight combinations

Download full size

Table 1. Evaluation index of abdomen slice test results from different methods

View table

Table 1. Evaluation index of abdomen slice test results from different methods

Energy /keV	Index	RICCS-SI^［22］	RED-CNN^［23］	DRN^［9］	TransCT^［24］	DTE-PDN
20‒80	PSNR /dB	33.9413	36.4773	37.0067	37.5022	37.8992
	SSIM	0.9692	0.9852	0.9893	0.9950	0.9958
	FSIM	0.9214	0.9515	0.9742	0.9963	0.9972
80‒100	PSNR /dB	34.3815	37.7927	38.0993	38.9681	39.2680
	SSIM	0.9703	0.9883	0.9891	0.9975	0.9964
	FSIM	0.9257	0.9543	0.9760	0.9969	0.9976
100‒120	PSNR /dB	35.0164	38.7362	39.0957	39.5153	39.6702
	SSIM	0.9755	0.9901	0.9933	0.9975	0.9977
	FSIM	0.9248	0.9556	0.9770	0.9971	0.9971

Table 2. Evaluation index of chest slice test results from different methods

View table

Table 2. Evaluation index of chest slice test results from different methods

Energy /keV	Index	RICCS-SI^［22］	RED-CNN^［23］	DRN^［9］	TransCT^［24］	DTE-PDN
20‒80	PSNR /dB	34.8861	36.9783	37.3461	37.5323	37.7301
	SSIM	0.9720	0.9873	0.9895	0.9927	0.9944
	FSIM	0.9311	0.9550	0.9762	0.9977	0.9976
80‒100	PSNR /dB	36.4913	38.2874	38.6820	39.6757	39.7969
	SSIM	0.9767	0.9892	0.9912	0.9956	0.9961
	FSIM	0.9368	0.9568	0.9774	0.9977	0.9981
100‒120	PSNR /dB	37.3751	39.2031	39.6868	39.8529	39.8615
	SSIM	0.9793	0.9906	0.9931	0.9981	0.9984
	FSIM	0.9325	0.9566	0.9771	0.9983	0.9984

Table 3. Comparison of the mean and standard deviation among ROIs in test results from different methods

View table

Table 3. Comparison of the mean and standard deviation among ROIs in test results from different methods

ROI	Energy /keV	RICCS-SI^［22］	RED-CNN^［23］	DRN^［9］	TransCT^［24］	DTE-PDN
ROI A	20‒80	0.2020±0.0042	0.2021±0.0030	0.2022±0.0027	0.2024±0.0019	0.2028±0.0017
	80‒100	0.1771±0.0034	0.1769±0.0023	0.1770±0.0020	0.1773±0.0015	0.1770±0.0015
	100‒120	0.1672±0.0030	0.1665±0.0021	0.1680±0.0018	0.1678±0.0014	0.1677±0.0013
ROI B	20‒80	0.2116±0.0043	0.2116±0.0031	0.2115±0.0032	0.2116±0.0028	0.2117±0.0028
	80‒100	0.1829±0.0036	0.1826±0.0027	0.1827±0.0026	0.1830±0.0023	0.1829±0.0022
	100‒120	0.1728±0.0033	0.1728±0.0026	0.1731±0.0024	0.1732±0.0021	0.1730±0.0020

Table 4. Evaluation index of abdomen and chest slices from DTE-PDN with different combination of blocks

View table

Table 4. Evaluation index of abdomen and chest slices from DTE-PDN with different combination of blocks

Slice	STU	LTU	Residual structure	20‒80 keV			80‒100 keV			100‒120 keV
Slice	STU	LTU	Residual structure	PSNR /dB	SSIM	FSIM	PSNR /dB	SSIM	FSIM	PSNR /dB	SSIM	FSIM
Abdomen	√			37.4853	0.9871	0.9796	38.8544	0.9847	0.9803	39.1781	0.9855	0.9814
		√		37.4965	0.9872	0.9908	38.9568	0.9879	0.9856	39.2517	0.9867	0.9823
	√	√		37.6030	0.9922	0.9953	39.1961	0.9931	0.9939	39.5180	0.9919	0.9948
	√	√	√	37.8992	0.9964	0.9972	39.2680	0.9964	0.9972	39.6702	0.9977	0.9971
Chest	√			37.3112	0.9861	0.9788	39.0217	0.9876	0.9790	39.1171	0.9873	0.9819
		√		37.4083	0.9873	0.9854	39.1445	0.9908	0.9835	39.2718	0.9895	0.9831
	√	√		37.4846	0.9907	0.9952	39.6001	0.9929	0.9953	39.6792	0.9923	0.9954
	√	√	√	37.7301	0.9944	0.9976	39.7969	0.9961	0.9981	39.8615	0.9984	0.9984

Tools

Get Citation

Copy Citation Text

Shunxin Ouyang, Zaifeng Shi, Fanning Kong, Lili Zhang, Qingjie Cao. Projection Domain Denoising Method for Multi-Energy Computed Tomography via Dual-Stream Transformer[J]. Laser & Optoelectronics Progress, 2024, 61(8): 0837008

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Digital Image Processing

Received: Jun. 1, 2023

Accepted: Aug. 8, 2023

Published Online: Mar. 13, 2024

The Author Email: Zaifeng Shi (shizaifeng@tju.edu.cn)

DOI:10.3788/LOP231426

CSTR:32186.14.LOP231426

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology