Automatic detection and diagnosis of diabetic retinopathy

Diabetes is a disease that affects millions of people not only in the United States (US), but worldwide. It affects the vasculature system throughout the entire body. There is a high concentration of blood vessels in the eye and the kidneys. When the blood vessels in the eye are affected the condition is referred as diabetic retinopathy (DR) and it has become the leading cause of blindness among adults in the US. The condition is treatable as long as it is detected in a timely manner. Diabetics need to schedule periodic eye examinations to prevent complications. In present health care systems screening is limited by its extensive need of human resources. An automated system for the detection and diagnosis of DR has been developed based on retinal image analysis, segmentation and understanding.; The automatic detection and diagnosis system can be divided into three main sections: normal retinal component detection, abnormality detection, and diagnostics. The normal retinal components of interest are retinal vasculature and fovea. The vasculature network provides a retinal map which is used in the detection of the fovea. Vessel detection is accomplished through a multi-resolution, multi-window parameter transform. The fovea contains a high concentration of photo detectors, thus providing the level of detail needed for reading and other everyday activities where sharp details are required. Abnormalities present in this area indicate a potential sight threatening condition. Fovea detection is accomplished via adaptive thresholding in intensity and probability space, geometrical features and location within the detected vessel network.; The abnormalities associated with DR vary in size as well as intensity. A filter bank combined with adaptive thresholding is used for abnormality segmentation. Feature extraction followed by a classifier is utilized for abnormality detection outside the fovea region which is more prone to false alarms.; The proposed diagnostics system consists of four grading levels: no DR, mild DR, moderate DR and severe DR. The levels are selected based on number, type, and location of the abnormalities referent to the fovea. The diagnostics system is implemented via an expert system based on rules published by the American Academy of Ophthalmology.