SmileAI is a cutting-edge project aimed at assisting early diagnosis and treatment planning in the field of oral healthcare through automated and accurate image segmentation. This project leverages the YOLOv11 large model to detect and classify various oral conditions.
SmileAI focuses on improving oral healthcare diagnosis using advanced object detection techniques. It uses the YOLOv11 large model trained on a dataset containing 2899 images annotated with various oral conditions, such as CALCULUS, Gingivitis, Lichen, Caries, DecayCavity, EarlyDecay, HealthyTooth, Missing, and Plaque.
This project can assist healthcare professionals by enabling early detection and better treatment planning.
- YOLO Version: YOLOv11
- Programming Language: Python 3.9
- Libraries: Numpy, Pandas, Roboflow, Ultralytics
- Environment: Jupyter Notebook
The dataset used in this project is sourced from RoboFlow Universe.
It consists of 2899 labeled images representing the following classes:
- CALCULUS
- Gingivitis
- Lichen
- Caries
- DecayCavity
- EarlyDecay
- HealthyTooth
- Missing
- Plaque
Make sure you have the following installed:
- Python 3.9
- Required libraries listed in
requirements.txt
- Clone the repository:
git clone https://github.com/tech-aakash/SmileAI-AI-Based-Diagnosis-for-Oral-Conditions.git cd SmileAI - Install the libraries
pip install -r requirements.txt
- Strong performance for Lichen (0.81) and DecayCavity (0.78).
- Misclassifications observed in Plaque and EarlyDecay, with spillover into "Background."
- Moderate accuracy for HealthyTooth (0.75), with some confusion with DecayCavity.
- Best F1 score: 0.54 at confidence 0.203.
- High F1 for Lichen and DecayCavity.
- Poor F1 for Plaque and Missing, indicating issues with these classes.
- High precision and recall for Lichen and DecayCavity.
- Low recall for Gingivitis and Plaque, indicating frequent missed detections.
- Overall mAP@0.5: 0.562.
- Loss metrics (box, classification, DFL) consistently decrease, indicating proper convergence.
- Validation loss slightly higher than training loss, suggesting mild overfitting.
- The model detects Plaque with a confidence score of 0.59. The bounding boxes indicate successful identification of the affected areas.
- Caries is detected with a confidence score of 0.78, showcasing the model's ability to identify decay accurately in detailed regions.
- The model detects Gingivitis with a confidence score of 0.26, which is relatively low, suggesting room for improvement in identifying inflammation conditions.