Development and Validation of an Automated Three-dimensional Cephalometry Method

Cephalometric analysis is a standardized diagnostic method used daily by orthodontists and maxillofacial surgeons. It is based on linear and angular measurements performed on radiographic images. This examination is traditionally done manually on two-dimensional radiographs, which does not allow to analyze finely the bilateral structures which are found superimposed. Cephalometric analysis of three-dimensional imaging (cone beam computed tomography (CBCT) or computed tomography scan (CT-Scan)) may provide additional diagnostic information, particularly for patients with maxillofacial abnormalities or marked asymmetries.

One of the obstacle to the clinical use of three-dimensional cephalometric analysis is the time and expertise needed to manually place the landmarks. Automatic methods described in literature are preliminary and lack validation in a clinical context.

Our retrospective observational study is aimed to develop and validate a new automated three-dimensional cephalometry method. This method will be based on a deep learning algorithm trained from a database of pre-surgery CT-Scans of patients with have undergone an orthognathic surgery. These CT-Scans will be manually annotated to provide a reference standard for the training of the algorithm and its evaluation. The validation of our results will focus on demonstrating the diagnostic effectiveness and robustness of three-dimensional cephalometric measurements obtained in this clinical context.