Analysis of Human Face Shape Abnormalities Using Machine Learning

Abstract

Because the capture of 3D facial form for people with medical conditions has become a practical reality, much research has been done using computer-based automatic methods to study 3D facial characteristics. This study focuses on face abnormality, especially, children with cleft lip and/or palate. Cleft lip is a birth defect that results in deformity of the upper lip and nose. Its severity is widely variable and the results of treatment are influenced by the initial deformity. Objective assessment of severity would help to guide prognosis and treatment. Given that facial asymmetry increases directly with increasing cleft severity, the main focus of this thesis is the quantification of asymmetry and nasal deformity. In our study, we developed an automated computer-based system for facial analysis that can greatly facilitate medical researchers. Our system takes raw 3D data, automatically crops the region of interest (the face), and normalizes the 3D face mesh. Then 20 landmarks are automatically located using geometric information followed by a deformable registration method. These landmarks are used for locating the mid-facial reference plane, which sets up the reference to compare the left and right side of the face. The plane is used to provide a ranking order based on the severity of cleft lip using a learning-to-rank algorithm. Last but not least, asymmetry and nasal deformity descriptors are calculated to provide quantitative scores. The main contributions of this work are: an automated methodology for cleaning raw data and pose normalization; an automated landmark location method; a machine learning technique that detect landmark-related regions; two original algorithms to find the mid-facial reference planes and their evaluation; features for ranking the severity of cleft lip; and six descriptors for asymmetry and nasal deformity, which are highly correlated to the ranking orders provided by medical expert.