Exploring Early Estimation of Treatment Needs for Age-Related Macular Degeneration Patients Using Easily Generated Features

Age-Related Macular Degeneration (AMD), a leading cause of vision loss in older adults, poses a significant challenge in developing personalized treatment strategies due to its diverse patterns of progression and outcomes. Optical Coherence Tomography (OCT), a non-invasive diagnostic tool, offers clinicians a valuable means of identifying various retinal abnormalities, including AMD, and tracking disease progression. The differentiation between healthy and AMD or diabetic retinopathy in OCT is less of a challenge for the current vision pretrained architectures. The current work explores the use of transfer learning for going from B-scans to volumes and sequence of volumes. The transfer is from a rather general task of classification between Normal, CNV and Drusen, to a specific task of recognizing early signs that indicate future progression of the disease. Different types of basic features and aggregation are combined with EfficientNetV2 architectures on two datasets: a dataset that includes volumes with annotations at B-scans level and volume level, respectively a dataset that includes 94 patients and their treatment history. The first two examinations from the latter dataset are considered for estimating the required administration of anti-VEGF injections. Our current experiments suggest that the modern pretrained architectures are able to capture not only disease presence but also elements relevant to disease progressions, all without the need for complex annotations such as the ones required in semantic segmentation. When the networks are able to identify aspects that were not explicitly trained to recognize them, the discovery of new biomarkers, a better understanding of the disease, and personalized treatments become more feasible.