Deep learning based data analysis and security in ophthalmology images

When ophthalmological diseases are not diagnosed and treated on time, results leading to blindness occur. Many studies have shown that early treatment can prevent these vision-threatening diseases. For example, diabetic retinopathy affects 80% of diabetic patients in the world and is the second biggest cause of blindness. Cataract, on the other hand, is generally an age-related disease and increases the blurring of vision over time, preventing the patient's vision. In this study, deep learning architectures were used for the detection of ophthalmological diseases. Thus, by developing automatic detection systems, it is aimed to speed up health services and to help specialists. Models created for this purpose must have high accuracy. Within the scope of the thesis, the detection of diabetic retinopathy and cataract from ophthalmological diseases was studied. To detect these ophthalmological diseases of the human retina; Models have been developed on diseased image datasets with methods such as image preprocessing, deep learning and transfer learning. With the developed models, high classification success has been achieved, which contributes to the literature. It was measured that the best model created for diabetic retinopathy reached 96.6% accuracy with 5 classifications, and the best model created for cataracts reached 97.2% accuracy with 2 classifications. The obtained accuracy rates contribute to the literature. In the analysis, it has been seen that the transfer learning method can make a better classification, at least 2%, than the classical deep learning methods. The best models created for each disease are made available in a web interface environment for use by experts. In the next step, the security requirements of the data collected on the web interface were taken into account. In this direction, the data stored on the server is encrypted with the most reliable algorithm specified in the literature, and patient data confidentiality is aimed. In this way, it is ensured that the data is stored reliably in the cyber environment.