The first steps within the 'Glaucoma in Focus' project have been taken. The project aims to deploy artificial intelligence (AI) to detect glaucoma early. A database of 100,000 images and associated findings is being used to train the AI algorithm.
Glaucoma in View is a project announced by Eye Hospital Rotterdam in March 2019 during International Glaucoma Week. It is the first large-scale project in Europe using AI to detect the eye disease glaucoma earlier via screenings.
Damaged optic nerve
Glaucoma is an eye disease in which the optic nerve is damaged. In many cases, this is due to increased pressure in the eyeball, causing the optic nerve to become pinched and die. This permanently damages the connection between the eye and the brain. In early glaucoma, images at the edges of the visual field disappear, something patients do not always realise immediately because their brains fill in the missing part of the image. Over time, more damage can occur and more image disappears, which can lead to socketed vision.
Large-scale screening for glaucoma is currently not readily available. AI could potentially offer a solution. An AI algorithm can fully automatically assess a photo of the inside of the eye - also known as a fundus photo. To make this possible, the algorithm is trained. Previously, similar projects have been carried out, in which AI algorithms managed to diagnose, among other things, lung photos of patients with COVBID-19 infection and some 50 eye diseases with high accuracy.
Database with 100,000 photos
Eye Hospital Rotterdam wants to train an AI algorithm to diagnose glaucoma. This will use a database of 100,000 eye photos and associated assessments. This database is made available by EyePACS, a US organisation that focuses, among other things, on screening for eye damage caused by diabetes.
Glaucoma specialist Dr Hans Lemij: "With EyePACS eye photos, we can start immediately. Taking photos takes a lot of time, especially since, due to the coronavirus, fewer patients enter our hospital and fewer photos are taken. I am therefore very happy with this cooperation!"
Reviewed by specialists
Each picture is assessed by at least two specialists, which include glaucoma specialists and trained optometrists. These specialists all go through a three-hour course and then take an exam, on which they must score a minimum of nine. Ten ophthalmologists have now passed the exam. In total, a team of 20 doctors will be deployed.
They will assess the photos via a website with a 'grading tool' developed by medical technology company DeepDee. Specialists can grade photos here by clicking on the 'glaucoma' or 'normal' button. If glaucoma is recognised, the specialist can explain his motivation for this. This information helps further refine the AI algorithm. Specialists are expected to start reviewing the imagery this summer.
The hospital expects its AI algorithm to learn enough from this database to be able to determine from a photo whether someone has glaucoma.
Benefits of AI
Using AI for such diagnoses offers great advantages. For instance, examinations can be performed faster, making it possible to perform such examinations on a larger scale. In addition, using AI reduces costs. It is also possible to provide people with the same quality diagnosis anywhere in the world. Eye Hospital Rotterdam estimates that the technology could prevent blindness and low vision in millions of people worldwide.
The project was funded through fundraising. Much of the money - over 500,000 euros - was raised at a benefit event at the Louwman Museum in 2019. Then-Minister for Medical Care Bruno Bruins called the project an example of innovation in healthcare where private funds are used for development.
Author: Wouter Hoeffnagel
