Optics and Precision Engineering, Volume. 30, Issue 17, 2147(2022)

Automatic classification of retinopathy with attention ConvNeXt

Wenbo HUANG1、*, Yuxiang HUANG1, Yuan YAO2, and Yang YAN1
Author Affiliations
  • 1School of Computer Science and Technology, Changchun Normal University, Changchun30032, China
  • 2Bureau of Major Tasks, Chinese Academy of Sciences, Beijing100864, China
  • show less
    Figures & Tables(10)
    Examples of APTOS 2019 blindness detection dataset sample
    Examples of cleaned up images
    Comparison of images before and after preprocessing
    Structure model of proposed ConvNeXt
    E-Block structure
    ECA module structure
    Classification confusion matrix of retinopathy for each model
    • Table 1. Comparison before and after dataset expansion

      View table
      View in Article

      Table 1. Comparison before and after dataset expansion

      数据类别扩充前扩充后
      0级-正常1 805幅2 019幅
      1级-轻度病变370幅2 035幅
      2级-中度病变999幅2 013幅
      3级-重度病变193幅1 982幅
      4级-增殖性DR295幅1 992幅
    • Table 2. ConvNeXt network structure details

      View table
      View in Article

      Table 2. ConvNeXt network structure details

      结 构输 入卷积核及步距输 出

      卷积层1

      E-Block层1

      224×224×34×4, s456×56×96
      56×56×96d7×7, s156×56×96
      56×56×961×1, s156×56×384
      56×56×3841×1, s156×56×96

      下采样

      E-Block层2

      56×56×962×2, s228×28×192
      28×28×192d7×7,s128×28×192
      28×28×1921×1,s128×28×768
      28×28×7681×1, s128×28×192

      下采样

      E-Block层3

      28×28×1922×2, s214×14×384
      14×14×384d7×7,s114×14×384
      14×14×3841×1, s114×14×1 536
      14×14×1 5361×1, s114×14×384

      下采样

      E-Block层4

      14×14×3842×2, s27×7×768
      7×7×768d7×7, s17×7×768
      7×7×7681×1, s17×7×3 072
      7×7×3 0721×1, s17×7×768
    • Table 3. Results of diabetic retinopathy grading by different algorithms

      View table
      View in Article

      Table 3. Results of diabetic retinopathy grading by different algorithms

      算 法灵敏度特异度准确率
      ConvNeXt(数据扩充前)75.8296.0685.11
      ConvNeXt94.2698.5694.27
      Densenet1591.7297.9491.72
      Shufflenet1691.7897.9691.77
      ConvNeXt+CBAM94.1898.5494.16
      ConvNeXt+SE94.5298.6494.51
      ConvNeXt+ECA95.2098.8095.21
    Tools

    Get Citation

    Copy Citation Text

    Wenbo HUANG, Yuxiang HUANG, Yuan YAO, Yang YAN. Automatic classification of retinopathy with attention ConvNeXt[J]. Optics and Precision Engineering, 2022, 30(17): 2147

    Download Citation

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

    Category: Information Sciences

    Received: May. 31, 2022

    Accepted: --

    Published Online: Oct. 20, 2022

    The Author Email: Wenbo HUANG (huangwenbo@sina.com)

    DOI:10.37188/OPE.20223017.2147

    Topics