DeepDent is a Deep Learning application designed to assist in identifying dental caries (decay) from Orthopantomogram (OPG) X-rays. By leveraging Transfer Learning with the MobileNetV2 architecture, the model provides high-accuracy screening to aid dental professionals in diagnostics.
The project follows a structured Machine Learning Lifecycle (MLLC):
- Data Engineering: Automated streaming of OPG X-ray datasets from Kaggle, followed by a controlled 80/20 split for training and validation.
- Preprocessing: Real-time Data Augmentation (rotation, zoom, flips) using
ImageDataGeneratorto improve model generalization. - Model Development: Implementation of Transfer Learning. We utilized a pre-trained MobileNetV2 base for feature extraction and added a custom Classification Head (GlobalAveragePooling, Dense, and Dropout layers).
- Training: Optimized using the Adam optimizer and Binary Cross-Entropy loss over 10 epochs.
- Deployment: Integration of the saved
.h5model into a responsive Streamlit web interface for real-time inference.
| Technology | Role |
|---|---|
| Python 3.13 | Core Programming Language |
| TensorFlow / Keras | Model Building & Training |
| MobileNetV2 | Pre-trained CNN Architecture |
| Streamlit | Web UI & Frontend Deployment |
| Kaggle API | Dataset Acquisition |
| Matplotlib | Training Visualization |
DeepDent_Train.ipynb: Jupyter Notebook containing the end-to-end training pipeline.app.py: Streamlit application file for the web interface.deepdent_model.h5: The serialized weights of the trained model.requirements.txt: List of dependencies for environment reproduction.CONTRIBUTOR_GUIDE.txt: Step-by-step instructions for setup and testing.
- Source: Dental OPG X-Ray Dataset
- Type: Panoramic X-rays (OPGs)
- Classes:
- Healthy: Structurally sound teeth.
- Decay: Visible dental caries/erosion.
git clone [https://github.com/RRonium/DeepDent.git](https://github.com/RRonium/DeepDent.git)
cd DeepDent
pip install -r requirements.txt