This project identifies teeth in X-ray images using instance segmentation and colored labeling. It originally used a PyTorch based Unet for the model but had an accuracy of less than 80%. As the main developer, to improve the project's performance, I modified the project by using a Res-Unet model instead and achieved 90% accuracy with a Dice Score of 91
This project was originally created by three computer science undergraduates from Boston College: Yuting Ji, Robert Smithers, and Danilla Zunic
Packages below are recommended to install.
- PyTorch
pip install torch
Teeth Labelling is not a brand new field of study, though the amount of available data is highly limited. Teeth data is rarely open source, a product of scarcity in related machine learning research as well as protection of patients' privacy.
With this teeth segmentation utility, artifical intelligence can autonomously label teeth scans and identify malformed, missing, or otherwise important concerns relating to the count and placement of teeth. The ability to label teeth in fractions of a second serves as a vital aide to dentistry personnel.
The written paper in CVPR format can be found at 'TeethSeg.pdf'. I hope these will help advance research within the academic community!
Due to privacy policy, I cannot provide the data for this model. However, the code is provided above in 'train.ipynb' and 'test.ipynb'. You may copy/reproduce a similar approach for your own x-ray teeth datasets.
- Paper
- To read the Paper, click on the PDF file: 'TeethSeg.pdf' - Code
- To view the main code, click the jupyter notebook named 'train.ipynb' - To test results, click the jupyter notebook named 'Test.ipynb'
Distributed under the MIT License. See LICENSE.txt for more information.
Yuting Ji - [email protected]
Original Project Link: https://github.com/RobertSmithers/TeethSegmentation