Early detection of diabetes retinopathy by new algorithms for automatic recognition of vascular changes

Diabetes mellitus often results in diabetic retinopathy caused by pathological changes of the retinal vessel tree. Early detection of these changes can delay the disease. Image processing can reduce the workload of screeners and can play a central role in quality assurance tasks. Therefore we aimed at the refinement and development of image processing algorithms to improve the quality and cost effectiveness of screening and diagnosis of diabetic retinopathy. In order to support ophthalmologists in their routine and to enable the quantitative assessment of vascular changes in colour fundus photographs a multi-resolution approach was developed which segments the vessel tree efficiently and precisely into digital images of the retina. The vessel tracker aims at determining as correctly as possible the retinal vascular network captured on a digital image irrespective of its origin. In addition to the tracker, algorithms were developed to detect the optic disk, bright lesions such as cotton wools spots, and dark lesions such as haemorrhages. The following classification of veins and arteries identifies arteries in 78.4 % and veins in 66.5% correctly. This helps selecting conspicuous images from a great number of patients.