Patient DR

During the DR screening, at least two images of each of the patient’s eyes are obtained and sent to the ophthalmologists for grading each eye separately. Both the images and the ophthalmologists’ diagnosis are stored in a Picture Archiving and Communication System (PACS).

Taking advantage of these large datasets, INESC TEC researchers have developed Patient DR, a weakly-supervised and multi-task method to train an image classifier using eye grades.

The software is able to use the information present in the current PACS to train an image classifier that is able to: 1) classify if an image depicts signs of DR; 2) classify if the patient should be treated; and 3) classify if an image has sufficient quality to be graded.