You can deploy this model and start earning money today!

yahia21/
Dental_AI_Classification21

Views

No views yet

🦷 Dental Disease Classification using Deep Learning

📋 Project Overview

Advanced deep learning model for classifying dental diseases from images using ResNet-50 architecture with balanced training techniques and regularization to prevent overfitting.

🎯 Supported Disease Classes

Calculus - Dental tartar buildup
Dental Caries - Tooth decay/cavities
Gingivitis - Gum inflammation
Hypodontia - Missing teeth (congenital)
Mouth Ulcer - Oral lesions
Tooth Discoloration - Abnormal tooth coloring

🚀 Model Performance

Metric	Score
Training Accuracy	94.17%
Validation Accuracy	90.83%
Balanced Accuracy	~90%
Overfitting Gap	3.34% ✅

✅ Low overfitting gap indicates excellent generalization!

📁 Project Structure

dental-classification/
├── data/
│   ├── train/           # Training images (organized by class folders)
│   ├── valid/           # Validation images
│   └── test/            # Test images
├── dental_classifier_balanced.pth    # 🔴 Trained model weights
├── train.py             # Training script
├── confusion_matrix_balanced.png     # Evaluation results
├── training_analysis_complete.png    # Training curves
└── README.md

📥 Model Download

Due to GitHub's file size limit (25MB), the trained model is hosted externally:

Download Link:

📦 Download dental_classifier_balanced.pth

File Size: ~XMB
SHA-256: (optional: add checksum here)

📍 How to Use:

Download the model file from the link above
Place it in the project root directory
Follow the inference instructions below

🛠️ Installation

Prerequisites

bash

Python 3.8+
CUDA-compatible GPU (recommended)

Install Dependencies

bash

pip install torch torchvision
pip install pillow numpy scikit-learn
pip install matplotlib seaborn tqdm

Or use:

pip install -r requirements.txt

🎓 Training the Model

Dataset Preparation

Organize your data as follows:

data/
├── train/
│   ├── Calculus/
│   ├── Data caries/
│   ├── Gingivitis/
│   ├── hypodontia/
│   ├── Mouth Ulcer/
│   └── Tooth Discoloration/
├── valid/ (same structure)
└── test/ (same structure)

Run Training

python train.py

Key Features:

⚖️ Balanced training with weighted loss
🔒 Strong regularization (dropout 0.6, weight decay 5e-4)
🧊 Frozen early layers to prevent overfitting
📉 Learning rate scheduling with ReduceLROnPlateau
⏱️ Early stopping (patience=7 epochs)
🚀 Mixed precision training (AMP) for GPU efficiency

🔮 Inference / Prediction

python

import torch
from PIL import Image
import torchvision.transforms as transforms

# Load model
checkpoint = torch.load('dental_classifier_balanced.pth')
model = RegularizedDentalClassifier(num_classes=6)
model.load_state_dict(checkpoint['model_state_dict'])
model.eval()

# Prepare image
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], 
                       std=[0.229, 0.224, 0.225])
])

image = Image.open('path/to/dental_image.jpg').convert('RGB')
input_tensor = transform(image).unsqueeze(0)

# Predict
with torch.no_grad():
    output = model(input_tensor)
    _, predicted = torch.max(output, 1)
    
class_names = checkpoint['class_names']
print(f"Predicted: {class_names[predicted.item()]}")

🏗️ Model Architecture

RegularizedDentalClassifier
├── Backbone: ResNet-50 (pretrained on ImageNet)
│   ├── Frozen layers: First 70% of parameters
│   └── Fine-tuned layers: Last 30 layers
│
└── Classifier Head:
    ├── Dropout (0.6)
    ├── Linear (2048 → 256)
    ├── ReLU + BatchNorm
    ├── Dropout (0.42)
    └── Linear (256 → 6 classes)

Total Parameters: ~25M
Trainable Parameters: ~8M

Key Observations:

✅ Smooth convergence without oscillations
✅ Validation loss closely tracks training loss
✅ No signs of overfitting (small gap between train/val)
✅ Balanced accuracy confirms good performance on all classes

⚙️ Hyperparameters

Parameter	Value
Base Model	ResNet-50 (pretrained)
Input Size	224×224×3
Batch Size	32
Optimizer	Adam
Learning Rate	0.0001
Weight Decay	5e-4
Dropout Rate	0.6
Epochs	15 (with early stopping)
Loss Function	Weighted Cross-Entropy

🧪 Handling Imbalanced Data

The project implements multiple strategies:

Class Weighting - Weighted loss function based on class frequencies
Undersampling/Oversampling - Balance validation set
Balanced Accuracy Metric - Better evaluation for imbalanced data
Per-class Monitoring - Track performance for minority classes

📝 Citation

If you use this model in your research, please cite:

bibtex

@software{dental_classifier_2025,
  author = {Your Name},
  title = {Dental Disease Classification using Deep Learning},
  year = {2025},
  url = {https://github.com/yourusername/dental-classification}
}

🤝 Contributing

Contributions are welcome! Please:

Fork the repository
Create a feature branch (git checkout -b feature/AmazingFeature)
Commit changes (git commit -m 'Add AmazingFeature')
Push to branch (git push origin feature/AmazingFeature)
Open a Pull Request

📜 License

This project is licensed under the MIT License - see LICENSE file for details.

🙏 Acknowledgments

ResNet-50 architecture by Microsoft Research
PyTorch team for the excellent framework
Dataset contributors (add source if applicable)

⚠️ Disclaimer

This model is for educational and research purposes only. It should NOT be used as a substitute for professional medical diagnosis. Always consult with qualified healthcare professionals for dental health concerns.

⭐ If you find this project useful, please give it a star! ⭐

yahia21/Dental_AI_Classification21