FIG-GAN: Fundus Image Generation via Deep Learning Based Generative Adversarial Network for AMD Disease Diagnosis

Globally, age-related macular degeneration (AMD) remains a main cause of irreversible vision loss. Recently, deep learning models have primarily focused on classifying fundus images for early detection of AMD progression. However, existing models rarely address the generation of future progression-aware fundus images, particularly when complete real longitudinal follow-up scans are unavailable. This limitation makes it difficult to track retinal changes over time and highlights the need for generative models capable of producing realistic drusen-level structural variations. To address these issues, a novel deep learning-based FIG-GAN model is to generate synthetic future fundus images from baseline inputs. Multi-Attention U-Net (MAU-Net) is initially employed to accurately segment drusen regions by integrating spatial, channel, and temporal attention mechanisms. In proposed FIG-GAN, the generator-1 uses elliptical sampling to smooth anatomical deformations and generator-2 uses concave-convex sampling to recreate advanced retinal changes related with AMD progression. The integration of these two sampling modules enable generation of realistic structural variations while preserving retinal integrity. Moreover, multi-scale discriminators are used to evaluate the generated retinal images at multiple resolutions that captures fine drusen-level texture variations. The quantitative evaluation demonstrates that the proposed FIG-GAN attains the highest accuracy of 98.12% and Fréchet inception distance of 12.84 ± 0.53 for fundus image generation. Furthermore, MAU-Net also attains strong segmentation performance with Jaccard index of 0.85 and Dice index of 0.93. The qualitative results indicate that the FIG-GAN produces high-fidelity synthetic fundus images that can support ophthalmologists by visualizing potential AMD progression, thereby aiding early diagnosis and treatment planning.