A-eye: Implementation of AI in eye disease detection

Abstract

To prevent vision loss from eye diseases like glaucoma and retinopathy it is important to diagnose them on time. When you discover the diseases too late the vision will be irreversibly damaged. For this reason, diabetic people are screened preventively. In the past, this happened every year but now only every three years, all to save costs. Artificial intelligence is a promising technique in the field of eye disease detection, it is able to detect based on just a fundus picture if asymptomatic eye diseases like glaucoma or retinopathy are developing. It has the potential to make diagnosing these diseases easier and more affordable. But while AI systems often perform good in labs, the introduction in the real medical world often fails. This project focuses on designing an AI service around an AI eye detection algorithm using a human-centered approach. The goal is to create an AI service that fits and will be adapted in the clinical workflow of the different medical specialists.
The project is done for the company Ksyos, this is a digital healthcare institute in the Netherlands. The service is built on top of their current ophthalmology service. The new AI service uses besides ophthalmologists and optometrists also opticians. Opticians will in the new service do the pre-diagnosis of eye diseases, they do this by taking a fundus picture and run it through an AI system. This will filter out a lot of people who do not need the expertise of an optometrist. In the current ophthalmology service, 90% of the people screened have healthy eyes. The AI service will not be just a screening service but will be a disease detection service, which means that the GP needs to count on the validity of the results of the system. To guaranty the quality, the AI system will split the patients into three categories, green for people with clearly healthy eyes, red for people who have a severe case of an eye disease, and orange for people who probably have an eye disease. The orange and red patients will be sent based on the severity of the disease and the kind of disease to an optometrist or an ophthalmologist. Explaining the results of the AI screening will not be the responsibility of the optician. The optician does not have the knowledge and the capabilities to explain the results calmly and clearly to the patient, Ksyos will take over this task from him. Ksyos will send a letter with the results to the patient that explains calmly and clearly what the next step in the diagnosis will be. While the optician does not have to interact with the AI system the optometrists does, he needs to understand why the AI system makes certain decisions. They need to know what the AI system has found and where. This way they can interpret the decision of the AI system and decide if they agree with this or not.