الفهرس 
SummaryOnly 14 pages are availabe for public view
 |  | from | 74 |  |  |
| | | from | 74 | | |
Abstract
Diabetic Retinopathy (DR) is one of the leading causes of blindness. It is a complication of diabetes, known as diabetic eye disease. Patients, who suffer from diabetes, face a problem of developing diabetic retinopathy. Changing in the blood’s sugar level may cause changes in retinal blood vessels, causing a growth of abnormal blood vessels on the surface of the retina. Early changes do not threaten central vision but developing DR is threatening the sight. Early detection of diabetic retinopathy assists The prevention of this the problem. This work proposed a method for detection and grading of diabetic retinopathy stages into normal, mild, moderate, severe, and Proliferative diabetic retinopathy (PDR) from retinal fundus images. In this thesis, we present two diabetic retinopathy grading systems. The first proposed technique includes three stages: data preprocessing, feature extraction, and classification, using ResNET-50 network. Transfer learning technique was applied deep learning model. The most benefits from the first proposed techniques were obtained using careful preprocessing (i.e., normalization and data augmentation), which achieved 86.67% classification accuracy. The second proposed technique includes three main stages: data augmentation, feature extraction, and classification, with using ResNet-50 or AlexNet networks. Transfer learning technique was applied for each deep learning model. The second proposed technique showed the ability to improve the classification accuracy over the first proposed system with an overall accuracy of 97.56% due to the different data augmentation step in second proposed technique achieved the best result. The experiments and results showed the privilege over the state-of-the art techniques of diabetic retinopathy systems, which applied on the same dataset. These results proved the promise of the proposed systems for grading diabetic retinopathy. The thesis comprises five chapters that can be summarized as follow: • Chapter (1) presents a general introduction to Diabetic Retinopathy (DR) problem. It explains briefly the motivations, problem statement, objectives, and proposed system. • Chapter (2) explains the previous related work for diabetic retinopathy detection and grading. • Chapter (3) describes two standard diabetic retinopathy dataset (IDRID, Messidor), two DR grading systems, with a detailed description of each system framework. • Chapter (4) explains the performed experiments and their discussions, with visual and quantitative results for both proposed grading DR systems, and includes a comparison the results of two proposed techniques. • Chapter (5) concludes the proposed study, future works, and the future plans.