Developing a Machine-Learning Algorithm to Diagnose Age-Related Macular Degeneration

Abstract

Age related macular degeneration (AMD) is an ocular disease that affects the central retina. The disease most commonly leads to blindness, and it involves a gradual deterioration of the retina cells or a fluid buildup in the retina that progressively hinders the sight. Ocular diseases including age-related macular degeneration are found in more than 196 million people aged 40 and up world wide, and these diseases are expected to grow to 240 million by 2050. The cost of the special machinery used to detect AMD can be a heavy burden for small eye clinics that are sporadically found in the regions outside of a city. As a result, we can find these advanced types of diagnostic machinery in inner city large hospitals and large eye clinics. Access to these types of machinery is a challenge that with affordable and efficient options will drastically lower the inequities regarding access to eye treatment. As a result of the issues regarding the diagnostic processes of age-related macular degeneration, a type of technology known as a convolutional neural network (CNN), a deep learning algorithm designed specially for images and pixel processing, can be especially used effectively in this scenario. A CNN completes this task in a fraction of the time it would take with conventional methods,  and is functional in use for other contexts as well, especially in areas where a comprehensive exam may not be feasible. Strategies that are needed to train CNN models for the assessment and diagnosis of ocular diseases have not yet been well characterized. By these ends, we trained a CNN model consisting of Convolution, Max Pooling, and ReLU layers on 5000 images of retinas affected by age-related macular degeneration, to conclude a diagnostic F1 score of 89%.