• Laser & Optoelectronics Progress
  • Vol. 57, Issue 4, 041013 (2020)
Hongyang Ruan1, Zhilan Chen2、*, Yingsheng Cheng3, and Kai Yang3
Author Affiliations
  • 1College of Engineering Science and Technology, Shanghai Ocean University, Shanghai 210306, China
  • 2College of Mechanical and Electrical Engineering, Shanghai Jian Qiao University, Shanghai 210306, China
  • 3Radiological Intervention Department, East Hospital of Shanghai Sixth People's Hospital, Shanghai 210306, China
  • show less
    DOI: 10.3788/LOP57.041013 Cite this Article Set citation alerts
    Hongyang Ruan, Zhilan Chen, Yingsheng Cheng, Kai Yang. Detection of Pulmonary Nodules Based on C-3D Deformable Convolutional Neural Network Model[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041013 Copy Citation Text show less
    CT images of lung
    Fig. 1. CT images of lung
    Structure of C-3D convolutional neural network model
    Fig. 2. Structure of C-3D convolutional neural network model
    Structures of 3D deformable convolution and pooling. (a) 3D deformable convolution; (b) 3D deformable pooling
    Fig. 3. Structures of 3D deformable convolution and pooling. (a) 3D deformable convolution; (b) 3D deformable pooling
    Structure of C-3D deformable convolutional neural network model
    Fig. 4. Structure of C-3D deformable convolutional neural network model
    Classification accuracy for different learning rates and optimization functions. (a) Experimental comparison of learning rate; (b) experimental comparison of optimization function
    Fig. 5. Classification accuracy for different learning rates and optimization functions. (a) Experimental comparison of learning rate; (b) experimental comparison of optimization function
    ROC curves and PRC curves of different models. (a) ROC curves; (b) PRC curves
    Fig. 6. ROC curves and PRC curves of different models. (a) ROC curves; (b) PRC curves
    Boxes of deformable convolution layers with different numbers of features. (a) Box of AUC; (b) box of F1; (c) box of P; (d) box of R
    Fig. 7. Boxes of deformable convolution layers with different numbers of features. (a) Box of AUC; (b) box of F1; (c) box of P; (d) box of R
    Visualization results of convolution window pixel sampling. (a) Original labeled lung images; (b) original C-3D convolutional neural network; (c) improved C-3D convolutional neural network
    Fig. 8. Visualization results of convolution window pixel sampling. (a) Original labeled lung images; (b) original C-3D convolutional neural network; (c) improved C-3D convolutional neural network
    LayerF (number of convolutionalfeature maps)×S (kernal size)StrideActivation functionOutput
    3D conv116×(3×3×3)1×1×1ReLU32×32×32
    3D conv264×(3×3×3)2×1×1ReLU16×32×32
    3D pool164×(1×2×2)1×2×216×16×16
    3D conv3128×(3×3×3)1×1×1ReLU16×16×16
    3D pool2128×(2×2×2)2×2×28×8×8
    3D conv4256×(3×3×3)1×1×1ReLU8×8×8
    3D pool3256×(2×2×2)2×2×24×4×4
    3D conv5512×(3×3×3)1×1×1ReLU4×4×4
    3D pool4512×(2×2×2)2×2×22×2×2
    3D conv664×(3×3×3)1×1×1ReLU1×1×1
    3D conv71×(1×1×1)1×1×1Sigmoid1×1×1
    Fc (fully connected layer)1
    Table 1. Setting of parameters of convolution layer and pooling layer in C-3D deformable convolutional neural network
    NumberNameNumber ofsamplesNumber ofpulmonarynodules
    1LIDC-IDIR8881186
    2East Hospital of ShanghaiSixth People's Hospital300346
    3Hospital of KanghuaHaining Zhejiang200212
    Table 2. Dataset information
    ModelAUCRPF1
    C-3D0.9422±0.01050.9125±0.01310.9335±0.01350.8846±0.0036
    C-3D±DC0.9513±0.01270.9228±0.01580.9387±0.01010.8978±0.0097
    C-3D±DP0.9326±0.02410.8997±0.01790.9254±0.01570.8797±0.0103
    C-3D±DCP0.9575±0.00980.9183±0.00960.9413±0.00750.9067±0.0058
    Table 3. Comparison of results of different convolutional neural network models
    ModelRPF1AUC
    C-3D+DCP,F:80.9131±0.01250.9226±0.00480.8948±0.01070.9394±0.0117
    C-3D±DCP,F:160.9183±0.00960.9413±0.00750.9067±0.00580.9575±0.0098
    C-3D±DCP,F:320.9253±0.01640.9534±0.01040.9286±0.00830.9617±0.0131
    C-3D±DCP,F:640.8957±0.00630.9113±0.02430.9024±0.00960.9339±0.0164
    Table 4. Results comparison of different output features in C-3D deformable convolutional neural network
    ModelAUCPRF1Data
    Ref. [6]0.84340.74440.79160.7673LIDC-IDRI
    Ref. [7]0.93870.87500.92430.8990LIDC-IDRI
    Ref. [8]0.85230.90450.91620.9103LIDC-IDRI
    Proposed method0.96170.95340.92390.9286LIDC-IDRI
    Proposed method0.96610.96130.92530.9378LIDC-IDIR+Local data
    Table 5. Comparison of different convolutional neural networks
    Hongyang Ruan, Zhilan Chen, Yingsheng Cheng, Kai Yang. Detection of Pulmonary Nodules Based on C-3D Deformable Convolutional Neural Network Model[J]. Laser & Optoelectronics Progress, 2020, 57(4): 041013
    Download Citation