Abstract:
Diabetic retinopathy (DR) is a significant eye disease that affects diabetic
patients and can lead to serious damage to the eye retina if it is not diagnosed and
treated in early stages. Early detection of this disease and quick intervention play a
crucial role in preventing severe eye damage and impairment. This thesis aims to
investigate the application of artificial intelligence (AI) techniques in detecting
diabetic retinopathy eye disease. The research focuses on developing a robust AI based system that can accurately identify and classify the stages of diabetic
retinopathy through analyzing fundus images of eye retinas. The proposed system
uses deep learning algorithms, specifically convolutional neural networks (CNNs),
which helps to extract relevant features from the images of eye retina and then
classify the images of diabetic retinopathy disease depending on the stage of the
disease in the eye. The performance of the developed detection system is evaluated
using a large dataset of 3662 retinal images, including images from patients with
various stages of diabetic retinopathy disease. The outcomes of this research include
an improved understanding of the potential of AI in the detection of diabetic
retinopathy and the development of a dependable and effective diagnostic tool to
assist medical professionals in early detection and diagnosis of this disease. The
results of this study have the potential to contribute significantly to the medical field,
especially Ophthalmology field, and lower the risk of vision loss among diabetic
patients. The proposed model is deep learning technique using a hybrid CNN-SVM
model, in which CNN is used for feature extraction and afterwards, SVM is used as a
classifier of diabetic retinopathy into 5 stages depending on the condition and
symptoms of the disease in the eye. The model, which was trained on Kaggle
APTOS 2019 dataset, reached a 96.23% testing accuracy, 96.2% sensitivity and
99.04% specificity, outperforming other classifiers and previously used models.