https://github.com/sanugiw/table-mounted-funduscope
The Table-Mounted Fundoscope is an advanced retinal imaging device designed to enhance the accuracy and efficiency of eye disease diagnosis. Unlike handheld devices, this stationary fundoscope minimizes motion-related distortions, providing clearer and more reliable images.
https://github.com/sanugiw/table-mounted-funduscope
ai-powered-analysis stable-imaging wifi-connection
Last synced: 2 days ago
JSON representation
The Table-Mounted Fundoscope is an advanced retinal imaging device designed to enhance the accuracy and efficiency of eye disease diagnosis. Unlike handheld devices, this stationary fundoscope minimizes motion-related distortions, providing clearer and more reliable images.
- Host: GitHub
- URL: https://github.com/sanugiw/table-mounted-funduscope
- Owner: Sanugiw
- Created: 2025-01-23T18:54:44.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2025-04-24T14:51:05.000Z (24 days ago)
- Last Synced: 2025-04-24T15:39:38.353Z (24 days ago)
- Topics: ai-powered-analysis, stable-imaging, wifi-connection
- Language: Jupyter Notebook
- Homepage:
- Size: 14.3 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# 👁️ ML-Integrated Fundoscope for Diabetic Retinopathy Detection
This project was developed as part of the **BM2210 – Biomedical Device Design** module (3rd semester) with the goal of advancing **automated eye disease diagnosis** using a smart, **AI-powered table-mounted fundoscope**. By combining **machine learning** and **IoT**, this system enables real-time, accessible, and accurate retinal imaging and classification of diabetic retinopathy severity levels.
---
## 🚀 Project Overview
We designed a **smart table-mounted fundoscope** using an **ESP32-CAM module** paired with a **20D lens** to capture high-resolution fundus images. The images are transmitted wirelessly over Wi-Fi for processing using a custom-trained machine learning model.
---
## 🔍 Key Features
### 🧠 Machine Learning Model
- **Trained on 21,000+ retinal images**
- Built using **TensorFlow** and **Keras**
- Utilizes a **Convolutional Neural Network (CNN)** to classify fundus images into 5 levels of diabetic retinopathy:
- **Class 0:** Normal Fundus
- **Class 1:** Mild Disease
- **Class 2:** Moderate Disease
- **Class 3:** Severe Disease
- **Class 4:** Proliferative Disease
- Optimized with **dropout regularization** and careful hyperparameter tuning for strong accuracy.
### 🌐 IoT Integration
- Images captured using the **ESP32-CAM** are transmitted over **Wi-Fi** for real-time diagnosis.
- Supports **remote access**, making it ideal for **telemedicine applications** and use in **resource-limited settings**.
### 💻 Streamlit UI
- Interactive **Streamlit dashboard** allows clinicians to:
- Upload fundus images
- Get real-time classification results
- View **confidence levels** for each prediction
- Access **reference information** for each disease level
---
## 🌟 Vision and Impact
This project addresses the increasing need for **early detection of diabetic retinopathy**, especially in regions with limited access to ophthalmologists. By integrating **AI** and **IoT** technologies, our device empowers healthcare providers with **faster, data-driven decisions**, improving patient outcomes and preventing vision loss.
---
## 🧪 Tech Stack
- **Hardware:** ESP32-CAM, 20D lens
- **Software:** Python, TensorFlow, Keras, Streamlit
- **Enclosure:** SolidWorks
- **Communication:** Wi-Fi (ESP32 to PC)
- **Dataset:** A publicly available retinal fundus image dataset (21,000+ samples)
---
## 📌 Future Enhancements
- Add support for mobile diagnostics
- Integrate data logging for patient monitoring over time
- Improve image preprocessing and segmentation for higher model accuracy
---