“EyeWeS: Weakly Supervised Pre-Trained Convolutional Neural Networks for Diabetic Retinopathy Detection” is the title of the scientific article that earned the Best Practical Paper Award at the MVA 2019 (International Conference on Machine Vision Applications) held in Tokyo, Japan.
Since 2011, the «Best Practical Paper Award » has been granted to the authors of outstanding articles about technology that is already in use or will soon be put into practice. In 2019, the research carried out by INESC TEC researchers Pedro Costa, Teresa Araújo, Guilherme Aresta, Adrian Galdran, Ana Maria Mendonça and Aurélio Campilho, and by Asim Smailagie (Carnegie Mellon University) was the winner of said distinction.
The award-winning article describes a method for detecting diabetic retinopathy through images of the eye’s fundus, thus being able to locate certain regions with lesions – despite being trained with images that only include global information on the presence or absence of the pathology. With this project, the team of researchers demonstrates that it is possible to convert a pre-trained convolutional neural network into a weakly supervised model, thus increasing its performance and efficiency.
Diabetic retinopathy (DR) is one of the main causes of blindness in the developed world, but the early detection of the pathology could easily prevent it. With the increasing number of diabetic patients, the use of automated systems for the detection of DR can play an important role in the screening of diabetic retinopathy.
The study carried out by INESC TEC researchers is part of the SCREEN-DR project, funded by FCT, and is carried out in partnership with the ARSN – Regional Health Administration (North), the University of Aveiro, and the Ophthalmology Services of the S. João University Hospital.
The researchers mentioned in this news piece are associated with INESC TEC and UP-FEUP.