🩺 EarCare AI Clinic

AI-Powered Ear Disease Detection System

🚀 An intelligent healthcare web application that detects ear diseases from otoscopic images using Deep Learning, Explainable AI, and LLM-based insights.

---

📌 Overview

EarCare AI Clinic is a full-stack AI-powered system designed to assist doctors and patients in diagnosing ear diseases quickly and accurately.

The system analyzes ear (otoscopic) images using a deep learning model and provides:

• Disease prediction
• Confidence score
• Visual explanation (Grad-CAM)
• Human-readable medical insights

---

🎯 One-Line Pitch

AI-powered ear disease detection system that combines deep learning, explainable AI, and real-time web deployment to assist doctors and improve patient diagnosis.

---

🧠 Key Features

• 🔍 AI-based Disease Detection
• 🧠 Explainable AI (Grad-CAM heatmaps)
• 📊 Confidence Score + Risk Analysis
• 💬 LLM-based Medical Explanation
• 📁 Patient History Tracking
• 🔐 JWT Authentication (Login/Signup/Logout)
• 📱 Modern Responsive UI (React + Tailwind)
• ⚡ Fast Prediction (<5 seconds)

---

⚙️ How It Works

1. User uploads ear image or captures via camera
2. Image is sent to backend (Flask API)
3. Deep Learning model (EfficientNet) processes image
4. Model outputs:
   • Predicted disease
   • Confidence score
5. Grad-CAM generates heatmap (focus region)
6. LLM generates easy-to-understand explanation
7. Results displayed on frontend dashboard

---

📊 Dataset

• Dataset: Otoscopic Ear Images
• Total Images: 3000
• Classes: 5 (Balanced dataset)
• Format: JPEG (.jpg)

Class	Images
Acute Otitis Media	600
Cerumen Impaction	600
Chronic Otitis Media	600
Myringosclerosis	600
Normal	600

---

🧪 Deep Learning Approach

• Model: EfficientNet (Transfer Learning)
• Framework: TensorFlow / Keras
• Image Size: 224 × 224

Techniques Used:

• Transfer Learning
• Data Augmentation
• Batch Training
• Hyperparameter Tuning

---

📈 Model Performance

• Initial Accuracy: ~10–15%
• Final Accuracy: ~98%
• Validation Accuracy: ~95–97%

High accuracy achieved due to balanced dataset, transfer learning, and optimized training pipeline.

---

🔍 Explainable AI

We use Grad-CAM to visualize model attention:

• Highlights important regions in the image
• Improves trust and interpretability
• Useful for medical validation

---

💬 LLM Integration

• Converts predictions into human-readable medical insights
• Provides:
  • Disease explanation
  • Risk level
  • Suggested next steps

---

🏗️ System Architecture

User → Upload Image → Flask API → EfficientNet Model → Prediction → Grad-CAM → LLM → Frontend Dashboard

---

🛠 Tech Stack

Frontend:

• React.js
• Tailwind CSS

Backend:

• Flask

Machine Learning:

• TensorFlow / Keras
• EfficientNet

Database:

• MongoDB

Authentication:

• JWT (Access + Refresh Tokens)

Extras:

• Grad-CAM
• LLM Integration

---

📸 Screenshots

Add your project screenshots here:

• UI Dashboard
• Image Upload Interface
• Prediction Output
• Grad-CAM Heatmap

---

🚀 Impact

• ⚡ Diagnosis in under 5 seconds
• 👨‍⚕️ Assists doctors (not replacement)
• 🧑‍🤝‍🧑 Easy for patients to understand
• 🌍 Useful in rural / low-resource areas

---

⚠️ Limitations

• Performance depends on image quality
• Not a replacement for professional diagnosis
• Requires further validation on real-world data

---

🔮 Future Work

• Mobile application deployment
• Integration with hospital systems
• Use of advanced models (ResNet, Vision Transformers)
• Larger and more diverse dataset

---

👨‍💻 Author

• SRV

---

📜 License

This project is for academic and research purposes only.

---

🙌 Acknowledgements

• Medical imaging datasets
• Deep learning research community
• Open-source tools and libraries

---

⭐ Show Your Support

If you found this project useful, consider giving it a ⭐ on GitHub!