🧠 Eye Disease Classification Using Deep Learning

This project focuses on classifying common eye diseases from retinal images using Convolutional Neural Networks (CNNs) with Transfer Learning. It is backed by a peer-reviewed publication and trained on publicly available datasets. The goal is to assist early diagnosis and decision-making for ophthalmologists using automated image analysis.

---

📄 Publication

📘 EYE DISEASE CLASSIFICATION USING DEEP LEARNING TECHNIQUES
📍 Industrial Engineering Journal
🗓️ Volume 51, Issue 11, No. 1, November 2022
📚 ISSN: 0970-2555

📄

---

🧬 Problem Statement

Given a retinal image, the model predicts whether the patient is diagnosed with one of the following conditions:

• Normal (Healthy)
• Diabetic Retinopathy
• Cataract
• Glaucoma

---

📂 Dataset

• 🔗 Kaggle Dataset: Eye Disease Classification
• Total of 4000 images, with approximately 1000 images per class
• Collected from public datasets: Ocular Recognition, HRF, DRIVE, and IDRiD
• Images are pre-labeled and suitable for supervised learning

---

🛠️ Model & Methodology

• Transfer Learning using pretrained CNN architectures:
  • ✅ ResNet50
  • ✅ VGG16
• Image augmentation applied:
  • 🔁 Rotation
  • 🔍 Zoom
  • 📏 Scaling
• Fine-tuned using:
  • SGD and Adam optimizers
  • Learning rate adjustments
  • Early stopping for regularization

---

🧪 Performance

• Validation set: 400 images (100 per class)
• Achieved Accuracy: 92%
• Trained using GPU acceleration on Kaggle
• Fine-tuning and hyperparameter optimization took ~1 week, with best training run completing in ~3 hours

---

🔧 How to Use

• 🔗 Model Training & Evaluation Notebook (Kaggle)
• The notebook contains all training, validation, and performance tracking steps
• Trained weights available (contact or download from Kaggle kernel output if published)

Note: This GitHub repository is under development and mirrors the Kaggle implementation.

---

📱 Future Work

We plan to develop a mobile application where:

• Users can upload a retinal image
• The app will predict the disease using the trained model
• Results can be used for clinical pre-screening

---

📃 License

To be added

---

👨‍💻 Author

Made with ❤️ by Gunavenkat Doddi
For academic inquiries or collaborations, feel free to reach out!

---

📌 Acknowledgements

Thanks to the creators of the public datasets and the Kaggle community. Special appreciation to the co-authors and reviewers of the published paper.