catarACTION: an Innovative Smartphone-based Tool to Detect Eye Disease

The Project

A staggering 2.2 billion people suffer from eye conditions and visual impairment, though over one billion of those cases could have been prevented or have still yet to be addressed–a casualty of widespread inaccessibility to hospitals and thus testing. In this project, I tackled this problem through an iOS app that takes retinal images with the help of a 3d-printed attachment. The app then saves this image for a machine-learning model, which analyzes the eye for a multitude of diseases before quickly giving the user their results. I built the app interface in Swift and used Firebase for storing user credentials and images. For the attachment, after testing several designs, I settled upon a direct ophthalmoscope-like one for its practicality and inexpensiveness. The total cost for it was $50, a fraction of the $6000+ price for slit lamps. In preparation for the machine-learning model, I compiled thousands of images and converted them into grayscale. I developed the model first on CreateML, though the validation and testing accuracies were low, around 76%. I migrated to Keras on Tensorflow instead to build a new model from scratch; the new validation accuracy was 91%, well above the medical benchmark of 80%. My project proved to be more affordable and just as accurate as traditional eye diagnostic measures. It can be used across the world on a massive scale.

View this project on YouTube