Aims: To develop a technique to detect microaneurysms automatically in 50 degrees digital red-free fundus photographs and evaluate its performance as a tool for screening diabetic patients for retinopathy.
Methods: Candidate microaneurysms are extracted, after the image has been modified to remove variations in background intensity, by algorithms that enhance small round features. Each microaneurysm candidate is then classified according to its intensity and size by the application of a set of rules derived from a training set of 102 images.
Results: When 3,783 individual images were analysed and the results compared with the opinion of a clinical research fellow examining the same images, the program achieved a sensitivity of 81% and a specificity of 93% for the detection of images containing microaneurysms. Nine hundred and twenty-five sets of 4 images per patient were then analysed and the total number of microaneurysms detected compared with the overall patient retinopathy grade derived by the clinician examining the same images. In this context, intended to mimic a screening situation, the program achieved a sensitivity of 85% and a specificity of 76% for the detection of patients with (any) retinopathy (positive predictive value 0.71, negative predictive value 0.88).
Conclusions: An automated technique was developed to detect retinopathy in digital red-free fundus images that can form part of a diabetic retinopathy screening programme. It is believed that it can perform a useful role in this context identifying images worthy of closer inspection or eliminating 50% or more of the screening population who have no retinopathy.