Age-related macular generation (AMD) — when left untreated — is the main cause of blindness for individuals over the age of 50. With the US population now counting over 100 million individuals over 50, it is imperative to develop methods that can effectively determine which individuals with an earlier, often asymptomatic stage, are at risk of developing the advanced stage that can cause severe vision loss. This paper studies the appropriateness of the transfer of image features computed from pre-trained deep neural networks to the problem in AMD detection. Tests using over 5600 images from the NIH AREDS dataset (the largest dataset used thus far for AMD image analysis studies) show good preliminary results (between nearly 92% and 95% accuracy).
