Journals Articles Conferences News About CLP
Submit a paper Email Alert
Search
Search
Articles
Journals
News
Advanced Search
Top Searches
2D MaterialsTransformation opticsQuantum PhotonicsSmall lasersSemiconductor UV PhotonicsSupersymmetric microring laser arrays

Chinese Journal of Lasers, Volume. 49, Issue 5, 0507208(2022)

Dual-Domain Neural Network for Sparse-View Photoacoustic Image Reconstruction

Kang Shen1,2, Songde Liu1,2, Junhui Shi3, and Chao Tian1,2、*
Author Affiliations
  • 1School of Engineering Science, University of Science and Technology of China, Hefei, Anhui 230026, China
  • 2Key Laboratory of Precision Scientific Instrumentation of Anhui Higher Education Institutes, Hefei, Anhui 230026, China
  • 3Zhejiang Lab, Hangzhou, Zhejiang 311121, China
    • show less
    AbstractGet PDF(in Chinese)
    Figures&Tables (11)
    Equations (0)
    References (49)
    Cited By (8)
    Tools
    Paper Information
    Topics
    Figures & Tables(11)
    Network structure of DI-Net. (a) Overall schematic of DI-Net; (b) network structure of D-Net (M＝512，N＝768，k＝16) and I-Net (M＝256，N＝256，k＝32)

    Fig. 1. Network structure of DI-Net. (a) Overall schematic of DI-Net; (b) network structure of D-Net (M＝512，N＝768，k＝16) and I-Net (M＝256，N＝256，k＝32)

    Download full size
    Schematic of the experimental setup

    Fig. 2. Schematic of the experimental setup

    Download full size
    Reconstruction results of vascular phantom based on 128 projection views (All color bars stand for amplitudes of pixels on images). (a) Reference image; (b)(d) images reconstructed by FBP algorithm, Post-Unet algorithm, and DI-Net algorithm, respectively; (e)(g) difference images between the reference image and the images reconstructed by FBP, Post-Unet, and DI-Net, respectively; (d) quantitative evaluation results of the reconstruction images

    Fig. 3. Reconstruction results of vascular phantom based on 128 projection views (All color bars stand for amplitudes of pixels on images). (a) Reference image; (b)(d) images reconstructed by FBP algorithm, Post-Unet algorithm, and DI-Net algorithm, respectively; (e)(g) difference images between the reference image and the images reconstructed by FBP, Post-Unet, and DI-Net, respectively; (d) quantitative evaluation results of the reconstruction images

    Download full size
    Reconstruction results of vascular phantom based on 256 projection views (All color bars stand for amplitudes of pixels on images). (a) Reference image; (b)(d) images reconstructed by FBP algorithm, Post-Unet algorithm, and DI-Net algorithm, respectively; (e)(g) difference images between the reference image and the images reconstructed by FBP, Post-Unet, and DI-Net, respectively; (d) quantitative evaluation results of the reconstruction images

    Fig. 4. Reconstruction results of vascular phantom based on 256 projection views (All color bars stand for amplitudes of pixels on images). (a) Reference image; (b)(d) images reconstructed by FBP algorithm, Post-Unet algorithm, and DI-Net algorithm, respectively; (e)(g) difference images between the reference image and the images reconstructed by FBP, Post-Unet, and DI-Net, respectively; (d) quantitative evaluation results of the reconstruction images

    Download full size
    Quantitative evaluation results of different algorithms on the vascular test dataset (To facilitate observation, the ordinate of the boxplot in the small dashed box is stretched and separately shown in the large dashed box). (a)(d) MSE; (b)(e) PSNR; (c)(f) SSIM

    Fig. 5. Quantitative evaluation results of different algorithms on the vascular test dataset (To facilitate observation, the ordinate of the boxplot in the small dashed box is stretched and separately shown in the large dashed box). (a)(d) MSE; (b)(e) PSNR; (c)(f) SSIM

    Download full size
    Reconstruction results of mouse slice based on 128 projection views (All color bars stand for amplitudes of pixels on images). (a) Reference image; (b)(d) images reconstructed by FBP algorithm, Post-Unet algorithm, and DI-Net algorithm, respectively; (e)(g) difference images between the reference image and the images reconstructed by FBP, Post-Unet, and DI-Net, respectively; (d) quantitative evaluation results of the reconstruction images

    Fig. 6. Reconstruction results of mouse slice based on 128 projection views (All color bars stand for amplitudes of pixels on images). (a) Reference image; (b)(d) images reconstructed by FBP algorithm, Post-Unet algorithm, and DI-Net algorithm, respectively; (e)(g) difference images between the reference image and the images reconstructed by FBP, Post-Unet, and DI-Net, respectively; (d) quantitative evaluation results of the reconstruction images

    Download full size
    Reconstruction results of mouse slice based on 256 projection views (All color bars stand for amplitudes of pixels on images). (a) Reference image; (b)(d) images reconstructed by FBP algorithm, Post-Unet algorithm, and DI-Net algorithm, respectively; (e)(g) difference images between the reference image and the images reconstructed by FBP, Post-Unet, and DI-Net, respectively; (d) quantitative evaluation results of the reconstruction images

    Fig. 7. Reconstruction results of mouse slice based on 256 projection views (All color bars stand for amplitudes of pixels on images). (a) Reference image; (b)(d) images reconstructed by FBP algorithm, Post-Unet algorithm, and DI-Net algorithm, respectively; (e)(g) difference images between the reference image and the images reconstructed by FBP, Post-Unet, and DI-Net, respectively; (d) quantitative evaluation results of the reconstruction images

    Download full size
    Quantitative evaluation results of different algorithms on the mouse slice test dataset. (a)(d) MSE; (b)(e) PSNR; (c)(f) SSIM

    Fig. 8. Quantitative evaluation results of different algorithms on the mouse slice test dataset. (a)(d) MSE; (b)(e) PSNR; (c)(f) SSIM

    Download full size

    • Table 1. Mean value of quantitative evaluation results for different algorithms on the vascular test dataset

      View table

      Table 1. Mean value of quantitative evaluation results for different algorithms on the vascular test dataset

      Number of viewsMethodMSEPSNR /dBSSIM
      128FBP1.880×10－223.180.4495
      Post-Unet5.005×10－439.050.9919
      DI-Net1.308×10－444.950.9974
      256FBP5.700×10－325.900.7463
      Post-Unet6.235×10－545.500.9978
      DI-Net3.640×10－547.820.9984

    • Table 2. Mean value of quantitative evaluation results for different algorithms on the mouse slice test dataset

      View table

      Table 2. Mean value of quantitative evaluation results for different algorithms on the mouse slice test dataset

      Number of viewsMethodMSEPSNR /dBSSIM
      128FBP0.084828.570.5385
      Post-Unet0.011937.000.8972
      DI-Net0.007239.260.9371
      2560.0218FBP33.770.7719
      Post-Unet0.004740.380.9462
      DI-Net0.002243.520.9741

    • Table 3. Comparisons of consuming time for different algorithms　unit：s

      View table

      Table 3. Comparisons of consuming time for different algorithms　unit：s

      Number of viewsFBPPost-UnetDI-Net
      1280.100.130.20
      2560.200.230.20
    Tools

    Get Citation

    Copy Citation Text

    Kang Shen, Songde Liu, Junhui Shi, Chao Tian. Dual-Domain Neural Network for Sparse-View Photoacoustic Image Reconstruction[J]. Chinese Journal of Lasers, 2022, 49(5): 0507208

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Set citation alerts for article
    Save article for my favorites
    Paper Information

    Received: Nov. 29, 2021

    Accepted: Jan. 12, 2022

    Published Online: Mar. 11, 2022

    The Author Email: Tian Chao (ctian@ustc.edu.cn)

    DOI:10.3788/CJL202249.0507208

    Topics

    laser devices and laser physics
    Lasers and Laser Optics
    Laser physics
    laser manufacturing
    Instrumentation, Measurement and Metrology