• Chinese Journal of Lasers
  • Vol. 50, Issue 21, 2107111 (2023)
Yixin Yuan1, Tao Chen2, Chengbo Liu2、**, and Jing Meng1、*
Author Affiliations
  • 1School of Computer, Qufu Normal University, Rizhao 276826, Shandong, China
  • 2Institute of Biomedical and Health Engineering, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, Guangdong, China
  • show less
    DOI: 10.3788/CJL230930 Cite this Article Set citation alerts
    Yixin Yuan, Tao Chen, Chengbo Liu, Jing Meng. Intelligent Skin-Removal for Photoacoustic Computed Tomography[J]. Chinese Journal of Lasers, 2023, 50(21): 2107111 Copy Citation Text show less
    PACT skin signal removal method based on deep learning
    Fig. 1. PACT skin signal removal method based on deep learning
    Flowchart of skin-integrity fitting. (a) Breaking in the middle of skin; (b) incomplete skin at the ends
    Fig. 2. Flowchart of skin-integrity fitting. (a) Breaking in the middle of skin; (b) incomplete skin at the ends
    Flowchart of mask generation and skin removing
    Fig. 3. Flowchart of mask generation and skin removing
    PACT imaging for human periphery blood vessels of legs. (a1)‒(f1) Original PACT images; (a2)‒(f2) ground truth; (a3)‒(f3) predicted results of Unet; (a4)‒(f4) predicted results of ResUnet; (a5)‒(f5) predicted results of MD-ResUnet
    Fig. 4. PACT imaging for human periphery blood vessels of legs. (a1)‒(f1) Original PACT images; (a2)‒(f2) ground truth; (a3)‒(f3) predicted results of Unet; (a4)‒(f4) predicted results of ResUnet; (a5)‒(f5) predicted results of MD-ResUnet
    Comparisons of different skin-removal methods. (a)(b) Original PACT MAP images from the two datasets; (c)(d) skin-removal MAP images with local weight fitting method; (e)(f) skin-removal MAP images with MD-ResUnet; (a1)‒(f1) typical B-Scan images and their enlarged sub-images
    Fig. 5. Comparisons of different skin-removal methods. (a)(b) Original PACT MAP images from the two datasets; (c)(d) skin-removal MAP images with local weight fitting method; (e)(f) skin-removal MAP images with MD-ResUnet; (a1)‒(f1) typical B-Scan images and their enlarged sub-images
    DatasetModelAccuracy /%Dice coefficient /%Specificity /%Sensitivity /%Average /%
    Dataset 1Unet80.392499.690099.873387.614191.8925
    ResUnet86.053699.700699.864088.937493.6389
    MD-ResUnet86.797299.707399.848090.429994.1956
    Dataset 2Unet69.764999.378299.719875.988286.2128
    ResUnet72.478899.369699.654979.834887.8345
    MD-ResUnet72.758599.362099.628381.124288.2183
    Table 1. Quantitative evaluation among different deep learning models
    DatasetMethodMSESSIM /%PSNR /dB
    Dataset 1Zhang et al.0.013698.665438.6774
    MD-ResUnet0.004099.047143.9745
    Dataset 2Zhang et al.0.007298.355738.4836
    MD-ResUnet0.003299.240341.9694
    Table 2. Quantitative comparison between traditional skin-removal method and proposed deep learning method
    Yixin Yuan, Tao Chen, Chengbo Liu, Jing Meng. Intelligent Skin-Removal for Photoacoustic Computed Tomography[J]. Chinese Journal of Lasers, 2023, 50(21): 2107111
    Download Citation