https://github.com/suryavamsi-p/diabetic-retinopathy-detection-with-resnet50

Built an end-to-end deep learning pipeline using ResNet-50 to classify retinal images into five stages of Diabetic Retinopathy. Applied transfer learning, image preprocessing, and AUC-based evaluation on the APTOS 2019 Kaggle dataset, achieving a 94% validation AUC—offering real-world potential in clinical diagnosis automation.
https://github.com/suryavamsi-p/diabetic-retinopathy-detection-with-resnet50

aptos2019 auc-evaluation clinical-decision-support cnn-model computer-vision data-augmentation deep-learning diabetic-retinopathy early-detection healthcare-ai image-classification keras medical-diagnosis medical-imaging multi-class-classification optical-imaging resnet50 retinal-image-analysis tensorflow transfer-learning

Last synced: 3 months ago
JSON representation

Built an end-to-end deep learning pipeline using ResNet-50 to classify retinal images into five stages of Diabetic Retinopathy. Applied transfer learning, image preprocessing, and AUC-based evaluation on the APTOS 2019 Kaggle dataset, achieving a 94% validation AUC—offering real-world potential in clinical diagnosis automation.

• Host: GitHub
• URL: https://github.com/suryavamsi-p/diabetic-retinopathy-detection-with-resnet50
• Owner: SuryaVamsi-P
• Created: 2023-10-09T22:38:46.000Z (about 2 years ago)
• Default Branch: main
• Last Pushed: 2025-05-24T21:25:51.000Z (5 months ago)
• Last Synced: 2025-06-12T19:07:31.620Z (4 months ago)
• Topics: aptos2019, auc-evaluation, clinical-decision-support, cnn-model, computer-vision, data-augmentation, deep-learning, diabetic-retinopathy, early-detection, healthcare-ai, image-classification, keras, medical-diagnosis, medical-imaging, multi-class-classification, optical-imaging, resnet50, retinal-image-analysis, tensorflow, transfer-learning
• Language: Python
• Homepage:
• Size: 2.15 MB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Diabetic Retinopathy Detection using ResNet-50

## Project Overview

This project presents a deep learning-based diagnostic solution for **early detection and classification of Diabetic Retinopathy (DR)**—a leading cause of vision loss among diabetic patients. Leveraging the power of **ResNet-50**, a pre-trained convolutional neural network (CNN), the model classifies high-resolution retinal images into five severity levels of DR:  

- 0 – No DR  

- 1 – Mild  

- 2 – Moderate  

- 3 – Severe  

- 4 – Proliferative DR  

Early detection of DR is crucial for timely medical intervention and preventing irreversible vision damage. This project serves as a decision-support tool for ophthalmologists and healthcare professionals by offering fast, automated classification of retinal images with strong accuracy.

## Key Highlights

- **Model Used:** ResNet-50 (pre-trained on ImageNet)  

- **Frameworks & Tools:** Python, TensorFlow, Keras, NumPy, Pandas, Matplotlib, Seaborn  

- **Techniques:** Image preprocessing, data augmentation, transfer learning, AUC & accuracy evaluation  

- **Performance:** Achieved **Training AUC: 97.77%**, **Validation AUC: 94%**  

- **Dataset:** APTOS 2019 Blindness Detection Dataset from Kaggle (35,000+ images)  

- **Problem Type:** Multi-class image classification (5 classes)

## Problem Statement

> Develop an automated deep learning model that can accurately detect and classify the stage of Diabetic Retinopathy from retinal fundus images, minimizing manual screening time and increasing diagnostic efficiency.

## Project Workflow

1. **Data Acquisition:** High-resolution retinal fundus images from Kaggle  

2. **Data Preprocessing:**  

   - Image resizing (224x224)  

   - Normalization and noise removal  

   - Data augmentation using `ImageDataGenerator`  

3. **Model Architecture:**  

   - Feature extraction with ResNet-50  

   - Fine-tuning the classifier head  

4. **Training:**  

   - Optimizer: Adam  

   - Loss: Categorical Crossentropy  

   - Epochs: 100  

5. **Evaluation:**  

   - Accuracy  

   - Loss  

   - AUC Score  

   - Confusion Matrix  

## Results

| Metric               | Value         |

|----------------------|---------------|

| Training Accuracy    | 87%           |

| Validation AUC Score | 94%           |

| Training AUC Score   | 97.77%        |

| Classification Type  | 5-class       |

Visual plots of accuracy, loss, and AUC are provided in the results section of the code.

## File Structure

```

.

├── DR.py                        # Main Python script with ResNet-50 implementation

├── trainLabels.csv             # Annotated labels for training

├── /Dataset/                   # Directory with retinal images

├── README.md                   # This file

└── [Additional plots & model artifacts]

```

## Mentor

This project was completed as part of the B.Tech final project at **Lakireddy Bali Reddy College of Engineering**, under the guidance of Mr. N. V. Naik, Sr. Assistant Professor.

**Author**

  

- **Surya vamsi Patiballa** [ MS in Data Science at George Washington University ]

  

- Email :- svamsi2002@gmail.com

- LinkedIn :- https://www.linkedin.com/in/surya-patiballa-b724851aa/

- Resume  :-  https://drive.google.com/file/d/178IYcArC6YYVdJiIwRmJYodzKZ-JXe-D/view?usp=sharing

## Future Enhancements

- Implement advanced models like **VGG16**, **InceptionV3**, or **EfficientNet**  

- Integrate hyperparameter tuning using **GridSearchCV**  

- Convert the model into a real-time **web-based diagnostic tool** using Flask/Django

## References

1. APTOS 2019 Blindness Detection – Kaggle Dataset :- (https://www.kaggle.com/competitions/aptos2019-blindness-detection)  

2. Deep learning papers cited in the included `Research Paper.pdf`