Implicit Neural Representations Model Geographic Atrophy Progression in AMD
A new research paper on arXiv (2608.10807) proposes using Implicit Neural Representations (INRs) to model the progression of Geographic Atrophy (GA) in Age-related Macular Degeneration (AMD), a leading cause of blindness in the Western world. The study addresses the challenge of highly individualized disease progression by generating both Fundus Autofluorescence (FAF) images and GA segmentations at past and future time points. The method achieves competitive segmentation quality, with the lowest Mean Absolute Error (MAE) for GA lesion area and the highest DICE score among comparison models, without sacrificing FAF image quality. The approach is designed for low-data settings, making it potentially useful for clinical applications. The code is available online.
Key facts
- arXiv:2608.10807
- Age-related Macular Degeneration (AMD) is a major cause of blindness in the Western world.
- Geographic Atrophy (GA) is the late dry phase of AMD, characterized by irreversible atrophic areas.
- Longitudinal Fundus Autofluorescence (FAF) imaging is the main tool for assessing lesion growth.
- The study proposes Implicit Neural Representations (INRs) to model GA progression.
- The method generates FAF and GA segmentation at past and future time points.
- The approach achieves the lowest Mean Absolute Error (MAE) for GA lesion area.
- The method achieves the highest DICE score among comparison models.
Entities
Institutions
- arXiv