Cardiff University | Prifysgol Caerdydd ORCA
Online Research @ Cardiff 
WelshClear Cookie - decide language by browser settings

3D face morphology classification for medical applications

Abbas, Hawraa 2018. 3D face morphology classification for medical applications. PhD Thesis, Cardiff University.
Item availability restricted.

[img]
Preview
PDF - Accepted Post-Print Version
Download (21MB) | Preview
[img] PDF - Supplemental Material
Restricted to Repository staff only

Download (165kB)

Abstract

Classification of facial morphology traits is an important problem for many medical applications, especially with regard to determining associations between facial morphological traits or facial abnormalities and genetic variants. A modern approach to the classification of facial characteristics(traits) is to use three-dimensional facial images. In clinical practice, classification is usually performed manually, which makes the process very tedious, time-consuming and prone to operator error. Also using simple landmark-to-landmark facial measurements may not accurately represent the underlying complex three-dimensional facial shape. This thesis presents the first automatic approach for classification and categorisation of facial morphological traits with application to lips and nose traits. It also introduces new 3D geodesic curvature features obtained along the geodesic paths between 3D facial anthropometric landmarks. These geometric features were used for lips and nose traits classification and categorisation. Finally, the influence of the discovered categories on the facial physical appearance are analysed using a new visualisation method in order to gain insight into suitability of categories for description of the underlying facial traits. The proposed approach was tested on the ALSPAC (Avon Longitudinal Study of Parents and Children) dataset consisting of 4747 3D full face meshes. The classification accuracy obtained using expert manual categories was not very high, in the region of 72%-79%, indicating that the manual categories may be unreliable. In an attempt to improve these accuracies,an automatic categorisation method was applied. In general,the classification accuracies based on the automatic lip categories were higher than those obtained using the manual categories by at least 8% and the automatic categories were found to be statistically more significant in the lip area than the manual categories. The same approach was used to categorise the nose traits, the result indicating that the proposed categorisation approach was capable of categorising any face morphological trait without the ground truth about its traits categories. Additionally, to test the robustness of the proposed features, they were used in a popular problem of gender classification and analysis. The results demonstrated superior classification accuracy to that of comparable methods. Finally, a discovery phase of a genome wide association analysis(GWAS) was carried out for 11 automatic lip and nose traits categories. As a result, statistically significant associations were found between four traits and six single nucleotide polymorphisms (SNPs). This is a very good result considering that for the 27 manual lip traits categories provided by medical expert, the associations were found between two traits and two SNPs only. This result testifies that the method proposed in this thesis for automatic categorisation of 3D facial morphology has a considerable potential for application to GWAS.

Item Type: Thesis (PhD)
Date Type: Submission
Status: Unpublished
Schools: Engineering
Uncontrolled Keywords: Face Morphology, Geodesic Path, Curvature, Machine Learning, 3D Imaging, ALSPAC datasets
Date of First Compliant Deposit: 4 May 2018
Last Modified: 04 May 2018 11:50
URI: http://orca.cf.ac.uk/id/eprint/111197

Actions (repository staff only)

Edit Item Edit Item

Downloads

Downloads per month over past year

View more statistics