eduzhai > Applied Sciences > Engineering >

Residual-CycleGAN based Camera Adaptation for Robust Diabetic Retinopathy Screening

  • king
  • (0) Download
  • 20210505
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: There are extensive researches focusing on automated diabetic reti-nopathy(DR) detection from fundus images. However, the accuracy drop is ob-served whenapplying these models in real-world DR screening, where the fun-dus camerabrands are different from the ones used to capture the training im-ages. Howcan we train a classification model on labeled fundus images ac-quired fromonly one camera brand, yet still achieves good performance on im-ages taken byother brands of cameras? In this paper, we quantitatively verify the impact offundus camera brands related domain shift on the performance of DRclassification models, from an experimental perspective. Further, we pro-posecamera-oriented residual-CycleGAN to mitigate the camera brand differ-ence bydomain adaptation and achieve increased classification performance on targetcamera images. Extensive ablation experiments on both the EyePACS da-taset anda private dataset show that the camera brand difference can signifi-cantlyimpact the classification performance and prove that our proposed meth-od caneffectively improve the model performance on the target domain. We haveinferred and labeled the camera brand for each image in the EyePACS da-tasetand will publicize the camera brand labels for further research on domainadaptation.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×