• Spectroscopy and Spectral Analysis
  • Vol. 41, Issue 9, 2950 (2021)
Xue-zhi REN1、*, Peng HE1、1; 2; *;, Zou-rong LONG1、1;, Xiao-dong GUO1、1;, Kang AN2、2;, Xiao-jie LÜ1、1;, Biao WEI1、1; 2;, and Peng FENG1、1; 2; *;
Author Affiliations
  • 11. Key Laboratory of Optoelectronics Technology & System (Chongqing University), Ministry of Education, Chongqing 400044, China
  • 22. ICT-NDT Engineering Research Center (Chongqing University), Ministry of Education, Chongqing 400044, China
  • show less
    DOI: 10.3964/j.issn.1000-0593(2021)09-2950-06 Cite this Article
    Xue-zhi REN, Peng HE, Zou-rong LONG, Xiao-dong GUO, Kang AN, Xiao-jie LÜ, Biao WEI, Peng FENG. Research on Spectral CT Image Denoising Via Fully Convolution Pyramid Residual Network[J]. Spectroscopy and Spectral Analysis, 2021, 41(9): 2950 Copy Citation Text show less
    FCPRN structure
    Fig. 1. FCPRN structure
    Spectral CT system based on photon-counting detector
    Fig. 2. Spectral CT system based on photon-counting detector
    Reconstructed spectral CT images of the six energy ranges at one positionThe energy ranges from (a) to (f) are 25~90, 30~90, 35~90, 40~90, 45~90 and 50~90 keV; The noise in reconstructed spectral CT images of different energy ranges are distinct
    Fig. 3. Reconstructed spectral CT images of the six energy ranges at one position
    The energy ranges from (a) to (f) are 25~90, 30~90, 35~90, 40~90, 45~90 and 50~90 keV; The noise in reconstructed spectral CT images of different energy ranges are distinct
    Reference images of the six energy ranges at one positionThe energy ranges from (a) to (f) are 25~90, 30~90, 35~90, 40~90, 45~90 and 50~90 keV
    Fig. 4. Reference images of the six energy ranges at one position
    The energy ranges from (a) to (f) are 25~90, 30~90, 35~90, 40~90, 45~90 and 50~90 keV
    Denoising results of DNCNN, REDCNN and FCPRN in three energy bins (25~90, 35~90 and 45~90 keV)
    Fig. 5. Denoising results of DNCNN, REDCNN and FCPRN in three energy bins (25~90, 35~90 and 45~90 keV)
    Details of denoising based on DNCNN, REDCNN and FCPRN in the three energy bins (25~90, 35~90 and 45~90 keV)
    Fig. 6. Details of denoising based on DNCNN, REDCNN and FCPRN in the three energy bins (25~90, 35~90 and 45~90 keV)
    LayersOut Dimension
    Input3
    3×3 Convolution48
    PR-blocks(8 layers)+TD(2 layers)64
    PR-blocks(8 layers)+TD(2 layers)80
    PR-blocks (8 layers)+TD(2 layers)96
    PR-blocks (8 layers)+TD(2 layers)112
    PR-blocks (8 layers)112
    PR-blocks (8 layers)112
    TU+PR-blocks (8 layers)96
    TU+PR-blocks (8 layers)80
    TU+PR-blocks (8 layers)64
    TU+PR-blocks (8 layers)48
    3×3Convolution3
    Table 1. The structure and parameters of FCPRN
    Energy/keVNetworksRMSEPSNRSSIM
    DNCNN0.008 341.663 10.980 1
    25~90REDCNN0.007 642.422 30.995 3
    FCPRN0.006 344.088 10.996 7
    DNCNN0.007 642.412 30.992 7
    30~90REDCNN0.007 242.915 90.992 9
    FCPRN0.005 944.576 80.993 2
    DNCNN0.007 542.541 20.979 3
    35~90REDCNN0.007 142.943 40.992 7
    FCPRN0.006 044.428 70.993 1
    DNCNN0.007 642.393 40.979 7
    40~90REDCNN0.007 342.743 80.992 3
    FCPRN0.006 144.212 30.995 7
    DNCNN0.007 842.180 20.979 8
    45~90REDCNN0.007 542.497 70.983 5
    FCPRN0.006 443.929 50.992 4
    DNCNN0.008 441.535 90.964 2
    50~90REDCNN0.008 141.802 90.983 1
    FCPRN0.007 143.035 60.990 0
    Table 2. Quantitative results of different networks
    Xue-zhi REN, Peng HE, Zou-rong LONG, Xiao-dong GUO, Kang AN, Xiao-jie LÜ, Biao WEI, Peng FENG. Research on Spectral CT Image Denoising Via Fully Convolution Pyramid Residual Network[J]. Spectroscopy and Spectral Analysis, 2021, 41(9): 2950
    Download Citation