Volume: 2 Issue: 2
Year: 2025, Page: 86-92, Doi: https://doi.org/10.70968/ijeaca.v2i2 .ML104
Received: July 28, 2025 Accepted: Nov. 30, 2025 Published: Dec. 23, 2025
Automated detection and localization of retinal pathologies from color fundus photographs is critical for early diagnosis of vision-threatening diseases. Existing supervised seg- mentation methods require expensive pixel-level annotations, while classification-only approaches lack spatial interpretability. This paper presents an enhanced framework for retinal disease detection and lesion localization by jointly optimizing a UVCGAN (UNet Vision Transformer Cycle-consistent GAN) generator with an EfficientNet-B0 classifier using only image-level labels. The key contribution is the replacement of the conventional Cycle- GAN generator with a UVCGAN generator that incorporates a Vision Transformer bottleneck within the U-Net encoder- decoder, enabling the model to capture global retinal structure through self-attention mechanisms. A PatchNCE contrastive loss enforces patch-level structural correspondence, improving upon CycleGAN’s cycle consistency approach. The generator translates diseased fundus images into healthy counterparts, and the resulting difference maps highlight lesion areas without pixel- level supervision. The EfficientNet-B0 classifier, jointly optimized with the generator, classifies difference maps while its gradient feedback simultaneously improves the generator’s lesion removal capability. Experimental evaluation on the ODIR-5K dataset with 832 images for pathologic myopia detection achieved 93.03% accuracy, 96.07% precision, 86.43% recall, 90.99% F1-score, and AUC of 0.984, demonstrating competitive performance on a realistically imbalanced dataset without pretrained classifier weights.
Keywords: Detection and Localization of Retinal Pathologies using Jointly Optimized UVCGAN and CNN Classifier on Color Fundus Photographs
1. Zhang Z, Ji Z, Chen Q, Yuan S, Fan W. Joint Optimization of CycleGAN and CNN Classifier for Detection and Localization of Retinal Pathologies on Color Fundus Photographs. IEEE Journal of Biomedical and Health Informatics. 2022; 26 (1). Available from: https://doi.org/10.1109/jbhi.2021.3092339
2. Torbunov D, Huang Y, Yu H, Huang J, Yoo S, Lin M, <I>et al</I>. UVCGAN: UNet Vision Transformer cycle-consistent GAN for unpaired image-to-image translation. 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV). 2023; Available from: https://doi.org/10.1109/wacv56688.2023.00077
3. Park T, Efros AA, Zhang R, Zhu JY. Contrastive Learning for Unpaired Image-to-Image Translation. Lecture Notes in Computer Science. 2020; Available from: https://doi.org/10.1007/978-3-030-58545-7_19
4. Zhu JY, Park T, Isola P, Efros AA. Unpaired image-to-image translation using cycle-consistent adversarial networks. 2017 IEEE International Conference on Computer Vision (ICCV). 2017; Available from: https://doi.org/10.1109/iccv.2017.244
5. Dosovitskiy A, <I>et al</I>. An image is worth 16x16 words: Transformers for image recognition at scale. <I>Proc. ICLR</I>, 2021.
6. He K, Zhang X, Ren S, Sun J. Deep Residual Learning for Image Recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2016; Available from: https://doi.org/10.1109/cvpr.2016.90
7. Tan M, Le QV. EfficientNet: Rethinking model scaling for convolutional neural networks. <I>Proc. ICML</I>, 2019, pp. 6105–6114.
8. Goodfellow I, <I>et al</I>. Generative adversarial nets. <I>Advances in Neural Information Processing Systems</I>, 2014, pp. 2672–2680.
9. Siddiquee MMR, Zhou Z, Tajbakhsh N, Feng R, Gotway M, Bengio Y, <I>et al</I>. Learning Fixed Points in Generative Adversarial Networks: From Image-to-Image Translation to Disease Detection and Localization. 2019 IEEE/CVF International Conference on Computer Vision (ICCV). 2019; Available from: https://doi.org/10.1109/iccv.2019.00028
10. Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation. Lecture Notes in Computer Science. 2015; Available from: https://doi.org/10.1007/978-3-319-24574-4_28
11. Isola P, Zhu JY, Zhou T, Efros AA. Image-to-Image Translation with Conditional Adversarial Networks. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). 2017; Available from: https://doi.org/10.1109/cvpr.2017.632
12. Peking University. <I>Ocular disease intelligent recognition (ODIR-5K)</I>. Kaggle Dataset, 2019.
13. Kingma DP, Ba J. Adam: A method for stochastic optimization. <I>Proc. ICLR</I>, 2015.
14. Heusel M, Ramsauer H, Unterthiner T, Nessler B, Hochreiter S. GANs trained by a two time-scale update rule converge to a local Nash equilibrium. NIPS'17: Proceedings of the 31st International Conference on Neural Information Processing Systems. 2017; Available from: https://doi.org/10.48550/arXiv.1706.08500
© 2025 Chaurasiya, et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Chaurasiya A, Jadhav H, Naik S, Rijhwani N, Chaudhari A. Detection and Localization of Retinal Pathologies using Jointly Optimized UVCGAN and CNN Classifier on Color FundusPhotographs. 2025; 2(2):86-92. https://doi.org/10.70968/ijeaca.v2i2.ML104