A novel deep learning method that makes automated screenings for eye diseases more efficient has been created by German researchers and is ready for use in clinics.
Researchers decided to look for ways to reduce the amount of annotated data needed to train an algorithm for accurate screening and diagnosis prediction.
They said a screening algorithm for diabetic retinopathy has been developed that needed 75% less annotated data and achieved the same diagnostic performance of human experts.
Researchers at Helmholtz Zentrum München, LMU University Eye Hospital Munich and the Technical University of Munich partnered on the project.
Using data from LMU, the algorithm was taught to predict retinal thickness, which enabled it to predict screening outcomes.
Dr Karsten Korteum, who led the clinical side of the study, said: ‘Automated detection and diagnosis of sight-impairing diabetic retinopathy with widely available fundus photography is a big improvement for screenings. Patient referrals to partly overcrowded specialised eye care centres could thus be reduced as well.’
Researchers added that the smaller algorithms could lead to deployment of the technology on mobile and embedded devices, and it could be applied to other eye diseases, such as age-related macular degeneration.