Decoding the maxillary morphological mechanism of mid-face depression or protrusion by an interpretable machine learning-assisted approach in an Asian female population.

For morphological assessments and treatment planning for the midface, the conventionally used measurement-based diagnosis is not potent enough to decode the complexity of craniofacial configuration. A more comprehensive assessment system using a machine learning-assisted approach is in need. In this study, the subnasale and maxilla positions were assessed in relation to the upper- and mid-facial structures in 1293 Asian females. Of these, 424 with proper mandibular and incisor positions were selected and classified into three types of facial appearance: balanced face, mid-face depression, and maxillary protrusion. Based on a comprehensive measurement system, machine-learning models were constructed to diagnose whether the patient had midfacial balance, deficiency, or excessive growth, in an interpretable manner. The XGBoost models performed the best for this task. The subnasale-to-glabella and maxilla-to-skull base anteroposterior relations were the most important features contributing to the model's diagnosis. The model still performed well without the input of facial angle, suggesting a capability of diagnosing irrespective of intermaxillary measurements. Running of the models on the external test sets revealed good efficacy in diagnosing mid-facial morphology in faces with proper mandibular positions, and in discriminating maxillary causes of facial aesthetic defects resulting from complex mechanisms, including skeletal and dental, and maxillary and mandibular abnormalities.