Timely diagnosis of craniofacial syndromes as well as adequate timing and choice of surgical technique are essential for proper care management. Statistical shape models and machine learning approaches are playing an increasing role in Medicine and have proven its usefulness. Frameworks that automate processes have become more popular. The use of 2D photographs for automated syndromic identification has shown its potential with the Face2Gene application. Yet, using 3D shape information without texture has not been studied in such depth. Moreover, the use of these models to understand shape change during growth and its applicability for surgical outcome measurements have not been analysed at length.
This thesis presents a framework using state-of-the-art machine learning and computer vision algorithms to explore possibilities for automated syndrome identification based on shape information only. The purpose of this was to enhance understanding of the natural development of the Apert syndromic face and its abnormality as compared to a normative group. An additional method was used to objectify changes as result of facial bipartition distraction, a common surgical correction technique, providing information on the successfulness and on inadequacies in terms of facial normalisation. Growth curves were constructed to further quantify facial abnormalities in Apert syndrome over time along with 3D shape models for intuitive visualisation of the shape variations. Post-operative models were built and compared with age-matched normative data to understand where normalisation is coming short.
The findings in this thesis provide markers for future translational research and may accelerate the adoption of the next generation diagnostics and surgical planning tools to further supplement the clinical decision-making process and ultimately to improve patients’ quality of life.