| Titre : | Diabetic retinopathy detection based on retinography images |
| Auteurs : | Afayez mokhtar BouhItem, Auteur ; Ahmed Aloui, Directeur de thèse |
| Type de document : | Monographie imprimée |
| Editeur : | Biskra [Algérie] : Faculté des Sciences Exactes et des Sciences de la Nature et de la Vie, Université Mohamed Khider, 2023 |
| Format : | 1vol.(98p.) / ill.couv.ill.encoul / 30 cm |
| Langues: | Anglais |
| Mots-clés: | Healthcare,Diabetes,Diabetic retinopathy, Artificial intelligence,Machine learning, Deep learning, CNN, DenseNet. |
| Résumé : |
The number of people with diabetes is expanding at an alarming rate throughout the world. According to the World Health Organization, there are currently 347 million cases of diabetes globally, and there will be 552 million cases by the year 2030. 1 in 3 of diabetes patients suffers from diabetic retinopathy, where DR is considered the fastest disease and the number one that leads to blindness because DR damages the retina’s blood vessels and DR is time sensitive side that require early detection to stop it from progressing to irreversible vision loss. Deep learning is one of the new and powerful solutions to detect and classify various diseases that are diagnosed by medical imaging like X-ray,MRI, and Fusndus imaging. The presented work it was with medical follow-up by an ophthalmologist surgeon in the eye hospital of Biskra. In our research, it is a deep learning methodology. We used the famous convolutional neural network AlexNet, a densely connected convolutional network DenseNet-121, and Yolov8. The two first models are used to classify the fundus images into five classes based on their severity: no DR, mild, moderate, severe, and porlifrate DR and the other model is used to detect the lesion of DR.We applied the three different models mentioned on different dataset with different structures. All used images are prepossessed with a Gaussian filter to improve the quality of the images and to emphasize the DR lesion like the blood vessels, microaneurysms and hemorrhages present in the retina. , also we used augmentation in the prepossessing phase and the circle crop method to give all images the circle retina form, and then we fed our data to our trainingmodels. Our results were very good and impressed the doctors at the hospital. |
| Sommaire : |
1 AI and healthcare 3 1.1 Introduction. 3 1.2 Healthcare . . 3 1.2.1 Definitions . .3 1.3 AI in Healthcare.... 4 1.4 Machine learning . . 7 1.4.1 Introduction .. 7 1.4.2 Types ofMachine learning algorithms . . 8 1.5 Deep learning . .. 10 1.5.1 Difference between DL and ML .. 11 1.5.2 Artificial neural networks . . 11 1.6 Convolutional Neural Networks . . 13 1.6.1 Definition . .. 13 1.6.2 Convolutional neural network composition and main operations . . 14 1.7 Popular CNN Architectures . 18 1.7.1 ResNet-50 . . 18 1.7.2 EfficientNet . 19 1.7.3 DenseNet . 20 1.8 Transfer learning . . 21 1.9 Conclusio.. 22 2 Diabetic retinopathy and related works .23 2.1 Introduction . . . 23 2.2 Diabetes . . .. 23 2.2.1 what is diabetes .. . 23 2.2.2 Types of diabetes . . 24 2.3 Diabetic retinopathy .. 25 2.4 Retinal imaging . . . 26 2.5 Diabetic retinopathy stages . 27 2.6 Related works .. 29 2.7 The critique of previous works . .32 2.7.1 Important note . 33 2.8 Conclusion . . 33 3 System design of a deep learning architecture for Diabetic Retinopathy detection 35 3.1 Introduction . . . 35 3.2 System design . . 36 3.2.1 General design . . 36 3.2.2 Detailed design . . 37 3.2.4 Preprocessing:. 43 3.2.5 Splitting dataset . . 44 3.2.6 Data Augmentation . 45 3.2.7 CNN Learning . . . 45 3.2.8 AlexNet . . 46 3.2.9 DenseNet-121 .. 47 3.2.10 YOLOv8 . . 49 3.2.11 Prediction . . . 51 3.2.12 Evaluation ofModels . . . 52 3.3 Conclusion . . 53 4 implementation and Results 54 4.1 Introduction . . 54 4.2 implementation of a Deep Learning architecture . . 54 4.2.1 Frameworks , tools and libraries . . . 54 4.2.2 Dataset preparation and preprocessing . . 58 4.2.3 Building ourModels . . 60 4.2.4 Training OurModels . . 67 4.3 Testing ourModel .. 70 4.3.1 Testing DenseNet and AlexNet . 70 4.3.2 Testing YOLOv8 . . . 72 4.4 Application interface . 72 4.4.1 Submission Interface . . 73 4.4.2 Results interface . . 4.5 First proposed Structure (Binary classification) . . . 74 4.5.1 The used Dataset . . . 74 4.5.2 AlexNet 75 4.5.3 Dens. . . 77 4.5.4 Comparisons of our experiments . .. 80 4.6 Second proposed Structure (Multi Classification) . . . 81 4.6.1 The used Dataset . . . 81 4.6.2 AlexNet . . . . 83 4.6.3 DenseNet-121 .. 4.6.4 Comparisons of our experiments . . .. 88 4.7 third proposed structure (Detection DR) . 89 4.7.1 The used Dataset . . 89 4.8 Comparison of our work with previous works . . 95 4.9 Conclusion . 98 |
| Type de document : | Mémoire master |
Disponibilité (1)
| Cote | Support | Localisation | Statut |
|---|---|---|---|
| Minf/819 | Mémoire master | bibliothèque sciences exactes | Consultable |
