Automatic cephalometric landmark identification in lateral skull X-Rays using convolutional neural networks

Cephalometric analysis is a study held in orthodontics, based on the identification of certain points in a skull image obtained through an X-ray image or another method in medical imaging. The indicated points are compared with standard values to evaluate and diagnose the patient. The radiograph’s labeling is regularly performed by hand, which makes the labeling process slow and prone to errors due to the visual acuity required. This approach is not much reproducible, because it relies on the domain and expertise of the expert labeler. Many machine learning methods were successfully applied to solve medical imaging tasks, aiming to reduce the health experts’ workload and emit more accurate diagnoses in less time and, avoid a more several clinical case. This work shows the design and development process of a machine learning system based on convolutional neural networks to identify 19 cephalometric landmarks for a lateral skull radiograph image as input. The system used a 400 labeled images dataset, from which, 150 were used for training, 150 for model’s validation and it was tested in the 100 remaining images.