Dental Cavity Detection

This project is a dental cavity detection system using YOLOv5. It is designed to analyze dental images and detect cavities using a trained deep learning model.

Features

• Real-time Detection: Detect cavities in dental images or live video streams.
• Custom Training: YOLOv5 is trained on a custom dataset of dental X-rays.
• High Accuracy: Uses a pretrained model fine-tuned for dental image analysis.

Requirements

Ensure the following are installed:

• Python 3.8 or above
• Torch (PyTorch)
• OpenCV
• YOLOv5
• Raspberry Pi (optional for embedded use)
• Camera Module (e.g., Pi Camera 3 Noir)

Python Dependencies

Install required Python libraries using: bash pip install -r requirements.txt

Example requirements.txt: torch opencv-python numpy matplotlib pandas seaborn PyYAML tqdm

Installation

1. Clone the Repository: bash git clone https://github.com/yourusername/dental-cavity-detection.git cd dental-cavity-detection

2. Download or Train YOLOv5 Weights:

   • Pretrained weights can be downloaded and placed in the weights directory.
   • Alternatively, train a YOLOv5 model on your dental dataset: bash python train.py --data data/dental.yaml --cfg models/yolov5s.yaml --weights yolov5s.pt --epochs 100

3. Connect a Camera (if using live detection):

   • Ensure your Raspberry Pi camera is enabled and recognized.

Usage

For Real-Time Detection:

Run the following command for live detection: bash libcamera-vid --camera 0 --width 640 --height 480 --framerate 30 --output - | python3 detect.py --source pipe:0 --weights weights/last.pt --img-size 640 --conf-thres 0.5 --view-img

For Image Detection:

Use this command to detect cavities in a static image: bash python detect.py --source data/images/sample.jpg --weights weights/last.pt --img-size 640 --conf-thres 0.5 --view-img

For Video Detection:

Run detection on a video file: bash python detect.py --source data/videos/sample.mp4 --weights weights/last.pt --img-size 640 --conf-thres 0.5 --view-img

Dataset

The dataset includes dental X-rays annotated for cavity detection. Ensure the dataset is in the correct YOLO format (images and .txt files for labels). Update the dataset path in data/dental.yaml: yaml train: ../data/dental/train val: ../data/dental/val test: ../data/dental/test

Results

Results will be saved in the runs/detect directory by default. Each detection run generates:

• Annotated images/videos with bounding boxes for cavities.
• Logs and performance metrics.

Troubleshooting

• If the camera is not detected, ensure libcamera is properly configured: bash libcamera-hello
• For OpenCV-related issues, verify its installation: bash pip install opencv-python

Future Improvements

• Add support for additional dental conditions.
• Optimize detection speed and model performance.
• Deploy the system on a cloud platform for scalability.

License

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

Acknowledgments

• YOLOv5 by Ultralytics
• OpenCV for image processing
• Dental dataset contributors