This project investigated the YOLO11n nano network, emphasizing its application for dental anatomy detection. The dataset originates from Roboflow. This type of transfer learning, implementing a pre-defined model on a new dataset, proved to be very useful and accurate. The model demonstrated strong performance with high accuracy in detecting tooth anatomy across a variety of images. The network effectively identified different tooth classes and localized them with high precision. 30 training epochs were enough for accurate results.
A few examples were performed and for each run we achieved the following evaluation scores:
| Example | Precision | Recall | F1 Score | mAP@50 | mAP@50:95 | Weighted Fitness |
Unweighted Fitness |
|---|---|---|---|---|---|---|---|
| 1 | 1.0000 | 0.9231 | 0.9600 | 1.0000 | 0.8417 | 0.9457 | 0.9412 |
| 2 | 0.9600 | 1.0000 | 0.9796 | 0.9600 | 0.7800 | 0.9329 | 0.9250 |
| 3 | 1.0000 | 1.0000 | 1.0000 | 1.0000 | 0.8286 | 0.9629 | 0.9571 |
| 4 | 0.9600 | 1.0000 | 0.9796 | 0.9600 | 0.7400 | 0.9264 | 0.9150 |
| 5 | 1.0000 | 1.0000 | 1.0000 | 1.0000 | 0.8417 | 0.9653 | 0.9604 |
Dental images with background overlays, including bounding boxes and confidence scores for each tooth type:
Overall, YOLO11 proves to be a reliable and efficient tool for dental image analysis, with its speed and performance making it suitable for real-time applications. Throughout all the performed examples, the YOLO network did not misrecognize any tooth labels.