Automated tooth crowding analysis

Deep learning to determine Little Irregularity Index on occlusal intra-oral photographs

Method

The method was developed with keypoint annotations on intra-oral scans and corresponding intra-oral photographs. The anatomic contact points were registered from the scan to the photo and YOLOv8 was used to determine tooth objects. Each tooth was predicted with a segmentation of its boundary, the mesial and distal anatomic contact points, and the physical size of the tooth in millimeters.

Installation

Install a Conda environment:

conda create -n crowding python=3.11
conda activate crowding

Install the Pip requirements:

pip install -r requirements.txt

Install Ultralytics YOLOv8

pip install -e ultralytics

Inference

You can run inference on the photograph in the figure above by running infer.py. Measurements of the anterior teeth will be saved to 'measurements.xlsx'. The model predictions can be visualized by running infer.py --verbose. Furthermore, the model can be run on your own photographs by specifying a folder with images using the in_dir argument.

Citation

@article{crowding_ai,
    author = {Hertig, Gabriel and van Nistelrooij, Niels and Schols, Jan and Xi, Tong and Vinayahalingam, Shankeeth and Patcas, Raphael},
    title = {Quantitative tooth crowding analysis in occlusal intra-oral photographs using a convolutional neural network},
    journal = {European Journal of Orthodontics},
    volume = {47},
    number = {3},
    year = {2025},
    month = {05},
    doi = {10.1093/ejo/cjaf025},
}