Vinayahalingam, S., Xi, T., Bergé, S., Maal, T., & de Jong, G. (2019). Automated detection of third molars and mandibular nerve by deep learning. Scientific Reports, 9(1), 9007.
We have 2 models that are part of the paper that can be loaded in-browser. The first model is obtaining a segmentation of the full dentition and the second model obtains subsequently the full dentition and the M3s (memory heavy). Only one model can be loaded at the time, refresh the page if you want to load the other models. We can’t convert the entire code of the paper to a browser version (yet). There is a fair chance it will not run on an older pc/laptop/tablet/etc. It is recommended to have at least 8Gb of RAM and optionally 4Gb of GPU RAM. See the paper for details on the implementation. You could apply for the code/models via the corresponding author of the paper.
These models can run offline. After successfully loading one of the models you can disable your internet connection and it should still work.
Only for research purposes! This may not work on your device! If you load a model you accept the Terms and Conditions!
- For details, meanings of abbreviations, etc. see the paper: https://doi.org/10.1038/s41598-019-45487-3
- Base Network: U-Net https://arxiv.org/abs/1511.06434
- Source: Can be applied for via the corresponding author
- Model files: Can be applied for via the corresponding author
- Input image source: 81 segmented OPGs with segmented dentition and M3’s at 1024x512px grayscale
- Training duration: ~up to 100 epochs, 4 samples/batch
- Online size:
- Dentition only: ~118Mb
- Lower M3s: 236 Mb
- GPU/CPU Free RAM Memory size (minimum):
- Dentition only: at least 1.4 Gb
- Lower M3s: at least 2.8 Gb
- Demo image: Case courtesy of Dr Andrew Dixon, Radiopaedia.org. From the case rID: 42336