Contexte scientifique

About the role

The goal of this internship is to contribute to the development of a predictive tool for estimating the risk of pulp exposure from preoperative dental radiographs. This tool aims to enable the least invasive, most predictable, and reproducible dental care procedures. The project leverages deep learning techniques, specifically neural networks, to segment dental radiographs and assess the risk of pulp exposure.

Untreated dental caries in permanent teeth is the most prevalent global disease, affecting 2.3 billion people (34% of the population) in 2017. While radiographic detection of deep carious lesions has high sensitivity and specificity, there is significant inter-evaluator variability in interpreting radiographs, particularly in estimating pulp proximity. This variability impacts treatment decisions, which range from minimally invasive caries removal to more invasive procedures like pulpotomy or endodontic treatment (root canal) in cases of pulp exposure.

Previous work by the supervising team found that directly predicting pulp exposure risk from radiographs is inconclusive. Thus, an initial segmentation step—automatically detecting healthy and carious areas—is essential for advanced radiographic analysis.

Responsibilities

• Understand the medical context of dental caries and radiographic imaging.
• Take ownership of the existing YOLOv8-based neural network for tooth detection and evaluate its performance.
• Train and evaluate a neural network for dental segmentation using a private dataset of 500 expert-annotated radiographs (including bounding boxes for teeth and segmented deep carious lesions).
• Use these networks to predict the risk of pulp exposure.
• Test the model’s generalizability on additional datasets, such as those from the DECAT clinical research project (ClinicalTrials.gov NCT 04607395).

Requirements

• Strong Python programming skills, with experience in PyTorch, NumPy, and SciPy.
• Knowledge of deep learning techniques and frameworks.
• Interest in medical imaging, particularly dental radiography.

Details

Duration: 4 to 6 months
Location: Laboratoire CREATIS and Faculté d’Odontologie, Lyon
Supervisors: Marie-Agnès Gasqui, Fabien Millioz, Matthieu Perard

Send your CV, cover letter, and latest transcript to: ma.gasqui@gmail.com, fabien.millioz@univ-lyon1.fr, matthieu.perard@univ-rennes.fr