Residual-CycleGAN based Camera Adaptation for Robust Diabetic Retinopathy Screening

There are extensive researches focusing on automated diabetic retinopathy (DR) detection from fundus images. However, the accuracy drop is observed when applying these models in real-world DR screening, where the fundus camera brands are different from the ones used to capture the training images. How can we train a classification model on labeled fundus images acquired from only one camera brand, yet still achieves good performance on images taken by other brands of cameras? In this paper, we quantitatively verify the impact of fundus camera brands related domain shift on the performance of DR classification models, from an experimental perspective. Further, we propose camera-oriented residual-CycleGAN to mitigate the camera brand difference by domain adaptation and achieve increased classification performance on target camera images. Extensive ablation experiments on both the EyePACS dataset and a private dataset show that the camera brand difference can significantly impact the classification performance and prove that our proposed method can effectively improve the model performance on the target domain. We have inferred and labeled the camera brand for each image in the EyePACS dataset and will publicize the camera brand labels for further research on domain adaptation.