Noise Reduction Model of Medical Ultrasound Images Based on Dual Attention Mechanism

Chenchen Xiong; Weili Jiang; Lizhong Jia; Dangguo Shao; Yan Xiang; Lei Ma; Jialin Yang

doi:10.3788/LOP202259.0217001

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 2 >Page 0217001 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 2, 0217001 (2022)

Noise Reduction Model of Medical Ultrasound Images Based on Dual Attention Mechanism

Chenchen Xiong¹, Weili Jiang², Lizhong Jia³, Dangguo Shao¹, Yan Xiang¹, Lei Ma^1、*, and Jialin Yang¹

Author Affiliations

¹Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming , Yunnan 650504, China

²College of Computer Science, Sichuan University, Chengdu , Sichuan 610065, China

³Medical Imaging Department, First Affiliated Hospital of Kunming Medical University, Kunming , Yunan 650504, China

show less

DOI: 10.3788/LOP202259.0217001 Cite this Article Set citation alerts

Chenchen Xiong, Weili Jiang, Lizhong Jia, Dangguo Shao, Yan Xiang, Lei Ma, Jialin Yang. Noise Reduction Model of Medical Ultrasound Images Based on Dual Attention Mechanism[J]. Laser & Optoelectronics Progress, 2022, 59(2): 0217001 Copy Citation Text

show less

Fig. 1. Schematic of our model

Download full size

Fig. 2. Schematic of channel attention

Download full size

Fig. 3. Effect comparison of simulated speckle noise images processed by several models

Download full size

Fig. 4. Effect comparison of liver ultrasound images processed by several models

Download full size

Fig. 5. Effect comparison of physical body membrane ultrasound images processed by several models

Download full size

Parameter	Content
Size of convolution kernel	（3，3）
The number of filters	64
The number of layers of DACNN	8
Ratio（compression rate of channel-attention）	8
Kernel（convolution kernel size of spatial-attention）	（7，7）
Loss function	Mean squared error（MSE）
Optimizer	Adam

Table 1. Experimental parameter setting

Model depth	PSNR	SSIM
2	24.21	0.82
3	25.47	0.81
4	23.37	0.81
5	25.07	0.82
6	25.26	0.82
7	25.45	0.83
8	25.82	0.82
9	25.28	0.82
10	24.77	0.82

Table 2. Effect of model depth on model performance

Image No.	PSNR					SSIM
Image No.	P-M	SRAD	DPAD	CNN	Our method	P-M	SRAD	DPAD	CNN	Our method
Mean	20.94	23.87	24.14	23.87	25.82	0.57	0.65	0.71	0.80	0.81
Figure 1	21.16	25.47	20.52	26.4	28.42	0.69	0.74	0.79	0.87	0.91
Figure 2	23.55	29.78	27.59	25.95	30.96	0.74	0.86	0.89	0.92	0.94
Figure 3	21.48	27.07	28.93	27.41	29.91	0.72	0.80	0.84	0.92	0.94
Figure 4	15.78	16.95	17.39	16.43	17.50	0.47	0.47	0.44	0.64	0.66
Figure 5	21.86	24.86	25.61	26.90	28.42	0.75	0.79	0.85	0.94	0.95
Figure 6	17.71	19.23	20.07	19.52	20.34	0.59	0.59	0.65	0.80	0.78
Figure 7	18.53	21.14	23.24	20.01	21.87	0.31	0.43	0.69	0.57	0.54
Figure 8	22.13	23.39	23.98	25.26	26.76	0.63	0.63	0.68	0.87	0.89
Figure 9	24.24	26.46	27.48	27.19	28.85	0.66	0.70	0.71	0.87	0.90
Figure 10	18.75	21.26	22.84	20.25	21.21	0.32	0.42	0.57	0.55	0.56
Figure 11	25.14	26.95	27.89	27.20	29.80	0.39	0.73	0.73	0.88	0.90

Table 3. Performance comparison of different models

Ultrasonic image	ENL					EPI
Ultrasonic image	P-M	SRAD	DPAD	CNN	Ours	P-M	SRAD	DPAD	CNN	Ours
Physical phantom	6.8060	9.2115	10.4624	8.5230	14.0220	0.3295	0.4231	0.3920	0.5270	0.5461
Liver	0.4420	0.4668	0.4750	0.5821	0.5983	0.6327	0.4752	0.3345	0.5558	0.5604

Table 4. Evaluation indexes comparison of real medical ultrasound images processed by several models

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information