Analysis of three-dimensional facial shape

Funding from the Wellcome Trust has been awarded to a consortium of institutions including the University of Glasgow to carry out research into the analysis of three-dimensional facial dsymorphology. The Face3D Project is the name given to this study and the aim is to develop methods of extracting information about the facial shape from three-dimensional images. There are numerous medical advantages in being able to accurately map the shape of a face in three-dimensions. This thesis is structured around the control facial shapes collected for the purpose of two medical applications; firstly the success of orthognathic surgery and secondly, the characterisation of the biological processes which underlie schizophrenia. This thesis covers the initial capture of a control data set of facial images and investigates methods of identifying and analysing information from the three-dimensional control facial shapes. In Chapter 2 the theory of various statistical methods used in shape analysis is described in detail. In Chapter 3 the stages of data collection are detailed including the recruitment process, exclusion criteria and image capture and data collection process itself. Chapter 4 details a subsequent validation study where the accuracy of image capture and the accuracy of anatomical landmark allocation is investigated. The design of the analysis takes the form of a four-layered hierarchical model where the source of variation in the model is calculated across each level. A measure of accuracy for individual landmark identification is found and an overall measure of accuracy in identifying a set of anatomical landmarks was found to be 1.81±0.84mm on average. In Chapter 5 the landmark configurations of all control faces are analysed and evidence of sexual dimorphism based on the shape of landmark coordinates after adjustment for scale is found. Males were found to have larger faces than females on average. The measure of participant’s body fat percentage on the position of anatomical landmarks is found to have a significant effect. In analysis of the landmarks on the midsagittal profile evidence of sexual dimorphism is found between the male and female landmark configurations after adjustment for scale. Subsequent Principal Component Analysis on the landmark configurations of males and females show little evidence of differences between males and females, despite formal analyses indicating evidence of sexual dimorphism. In Chapter 6 a method of curve identification is detailed for the philtrum ridges on the upper lip. Evidence of sexual dimorphism for the curves on the upper lip ridges was not identified. A general method of curve identification is then established. Curve identification of the midsagittal profile is described in detail. A formal test indicates that evidence sexual dimorphism is present in the shape of the midsagittal curves identified, however the result is borderline significant. After adjustment for body fat no difference is found between the mean shape of the midsagittal curves for males and females