README.md

Back to Projects List

Multi-stage dental segmentation using MONAI Label

Key Investigators

• David García Mato (Ebatinca S.L., Las Palmas de Gran Canaria, Spain)
• Yucheng Tang (NVIDIA, USA)
• Andres Diaz Pinto (NVIDIA, UK)
• Daniel Palkovics (Semmelweis University, Budapest, Hungary)
• Csaba Pinter (Ebatinca S.L., Las Palmas de Gran Canaria, Spain)
• Attila Nagy (University of Szeged, Szeged, Hungary)
• Brianna Burton (3D Side, Belgium)
• Umang Pandey (Universidad Carlos III de Madrid, Spain)

Project Description

A three-dimensional visualization of dento-alveolar structures can enhance the surgical planning process. However, no fully automated segmentation methods are currently available to generate realistic 3D virtual models of teeth, inferior alveolar nerves and alveolar bone.

Example: manual segmentation of teeth and alveolar bone.

We have already tested segmentation and deepedit models in MONAI Label. Those models are good for single-label teeth segmentation or mandible segmentation. However, results are not optimal when trying to perform a multi-label segmentation where all teeth are correctly identified and segmented.

Objective

1. Create MONAI pipeline for automatic segmentation of dental structures: teeth, mandible and inferior alveolar nerves.

Approach and Plan

1. Implement multi-stage approach for teeth segmentation using MONAI Label pipelines. At least, two stages: (1) teeth localization and (2) teeth segmentation.
2. Develop model to segment mandible and inferior alveolar nerves.
3. Combine multi-stage teeth segmentation with mandible and nerve segmentation.

Progress and Next Steps

1. Test Deepedit model for automatic segmentation of mandible and inferior alveolar nerves.

2. Create and test pipeline for dental segmentation using only two stages: teeth localization + teeth segmentation

3. Create and test pipeline for dental segmentation using three stages: teeth localization, teeth centroid computation and teeth segmentation. Pipeline based on vertebra pipeline with some modifications.

Illustrations

Result: teeth segmentation using dental pipeline (2 stages)

Result: mandible and inferior alveolar nerves segmentation using DeepEdit.

Background and References

• Related project from 37th NA-MIC Project Week: Multi-stage deep learning segmentation of teeth