Diabetic Retinopathy (DR) is a disease that affects up to 80% of diabetics around the world. It is the second greatest cause of blindness in the Western world, and one of the leading causes of blindness in the U.S. Many studies have demonstrated that early treatment can reduce the number of sight-threatening DR cases, mitigating the medical and economic impact of the disease. Accurate, early detection of eye disease is important because of its potential to reduce rates of blindness worldwide. Retinal photography for DR has been promoted for decades for its utility in both disease screening and clinical research studies. In recent years, several research centers have presented systems to detect pathology in retinal images. However, these approaches apply specialized algorithms to detect specific types of lesion in the retina. In order to detect multiple lesions, these systems generally implement multiple algorithms. Furthermore, some of these studies evaluate their algorithms on a single dataset, thus avoiding potential problems associated with the differences in fundus imaging devices, such as camera resolution. These methodologies primarily employ bottom-up approaches, in which the accurate segmentation of all the lesions in the retina is the basis for correct determination. A disadvantage of bottom-up approaches is that they rely on the accurate segmentation of all lesions in order to measure performance. On the other hand, top-down approaches do not depend on the segmentation of specific lesions. Thus, top-down methods can potentially detect abnormalities not explicitly used in their training phase. A disadvantage of these methods is that they cannot identify specific pathologies and require large datasets to build their training models. In this dissertation, I merged the advantages of the top-down and bottom-up approaches to detect DR with high accuracy. First, I developed an algorithm based on a top-down approach to detect abnormalities in the retina due to DR. By doing so, I was able to evaluate DR pathologies other than microaneurysms and exudates, which are the main focus of most current approaches. In addition, I demonstrated good generalization capacity of this algorithm by applying it to other eye diseases, such as age-related macular degeneration. Due to the fact that high accuracy is required for sight-threatening conditions, I developed two bottom-up approaches, since it has been proven that bottom-up approaches produce more accurate results than top-down approaches for particular structures. Consequently, I developed an algorithm to detect exudates in the macula. The presence of this pathology is considered to be a surrogate for clinical significant macular edema (CSME), a sight-threatening condition of DR. The analysis of the optic disc is usually not taken into account in DR screening systems. However, there is a pathology called neovascularization that is present in advanced stages of DR, making its detection of crucial clinical importance. In order to address this problem, I developed an algorithm to detect neovascularization in the optic disc. These algorithms are based on amplitude-modulation and frequency-modulation (AM-FM) representations, morphological image processing methods, and classification algorithms. The methods were tested on a diverse set of large databases and are considered to be the state-of the art in this field.
Diabetic retinopathy--Imaging--Data processing., Fundus oculi--Imaging--Data processing., Neovascularization--Imaging--Data processing., Image processing--Digital techniques.
Level of Degree
Electrical and Computer Engineering
First Committee Member (Chair)
Second Committee Member
Third Committee Member
Agurto Rios, Carla Paola. "Detection and Classification of Diabetic Retinopathy Pathologies in Fundus Images." (2013). https://digitalrepository.unm.edu/ece_etds/5