Machine (Deep) Learning for Characterization of Craniofacial Variations

The morphology of the craniofacial structure often varies in the general population, but the severities of the variation from the normal physiological structure can reveal potential disorders that affect the patient’s quality of life. In recent years, the preferred method for diagnosis and treatment of patients with craniofacial disorders has been using Cone Beam Computed Tomography (CBCT) imaging accompanied by manual segmentation to produce a 3D model of the craniofacial region. Unfortunately, manual segmentation is often tedious, user-dependent, and costly. The field of machine learning has flourished in recent years due to improvements in computer processing power, as well as storage capacity. Machine learning in different areas of the medical field has also been up-and-coming. One beneficial application of machine learning is automatic volumetric segmentation. This learning based method is much faster and can be even more accurate than manual segmentation. This chapter will first introduce the process of automated segmentation in detail, from preparing annotated data to developing a neural network. Then we will look at an application on a specific craniofacial variation, patients with unilaterally impacted canines. The review illustrates the feasibility and benefit of using machine learning in orthodontic treatment.