General Cephalometric Landmark Detection for Different Source of X-Ray Images

Cephalometric landmarks are specific anatomical points or structures identifiable on a radiograph or X-ray image of the human skull. These landmarks serve as reference points for cephalometric analysis, which involves measuring and assessing skull structures for various clinical and research purposes. Carrying out this marking can lead to errors, such as spending excessive time. Currently, implementations that involve machine learning help to mark these landmarks automatically. However, many of these need to consider that X-Ray images can be obtained from different machines, removing homogeneity in the observations. In this work, we study the influence of cephalometric images from different X-Ray machines on the performance of a model to detect cephalometric landmarks. Performance-based on age and the number of points to be estimated was also studied. We use a universal landmark detection model, a well-known architecture in the state of the art. The results showed that combining specific machines can achieve more minor errors and better generalization to machines not included in the training. On the other hand, the model can train with missing landmarks, but it can only succeed if the number of missing points is low.