AI algorithm improves glaucoma diagnosis – ICT&health

Glaucoma in itself is a fairly treatable condition. For this it is important that the eye disease is detected at an early stage. That is still a problem, and not just in developing countries. In the Netherlands, too, the correct diagnosis is made and appropriate treatment is initiated in only half of the cases. Failure to diagnose this eye disease in time can lead to irreversible vision loss and even blindness.

Better glaucoma diagnosis thanks to AI

For the development of the AI ​​algorithm first started populating a database with the fundus photos of more than 110,000 people. The project, ‘Glaucoma in Pictures’ started in 2020 and its main goal was to explore the possibility of using artificial intelligence (AI) in the assessment of eye images and the early diagnosis of glaucoma.

Those photos were assessed by 20 optometrists/ophthalmologists from various countries. All pictures were labeled ‘glaucoma’ or ‘no glaucoma’. This database was then used to ‘train’ the developed AI algorithm to diagnose this eye disease.

AI-challenge

At the beginning of this year, a so-called AI challenge was organized in which, in collaboration with the Artificial Intelligence department of the University of Amsterdam, developers were challenged to build an algorithm that can diagnose glaucoma at least as well, and preferably better, as the specialists. The ‘winning’ algorithm eventually achieved an 85% diagnosis rate for recognizing glaucoma, regardless of ethnicity and regardless of the fundus camera used. That’s just as good as a specialist would do.

Research has now shown that the algorithm can be used in a broad and easily accessible manner. With the help of AI, the chance of large-scale screening for this eye disease is greatly increased. The aim is to prevent and reduce blindness and impaired vision worldwide as much as possible thanks to early diagnosis and treatment.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.