Parents' perspectives and performance evaluation of facial analysis technologies for the diagnosis of congenital disorders

Abstract

Congenital disorders are a major health care burden. Most congenital disorders that are due to genetic causes do not have a cure, but an early and accurate diagnosis may alleviate associated symptoms and contribute to the correct management of the disorder. However, there is a lack of medical geneticists and doctors who can make these diagnoses in developing countries. Thus, facial analysis technologies can provide a quick and objective way to initially diagnose individuals with a congenital disorder where resources are limited because almost half of all inherited disorders have a typical facial gestalt. Chapter 1 is a literature review, focusing on facial analysis technologies and how it is used to make an initial diagnosis based on the typical facial features of an individual, with a special focus on Face2Gene. I briefly reviewed the four disorders under investigation in this study, their prevalence, cause, and particularly the typical facial features associated with each disorder. We first aimed to better understand parents’ views on the collection, storage, use, and publication of their children’s facial images for research and diagnosis. Large datasets of facial photographs are required to train facial analysis algorithms, and we wanted to better understand the public’s views on this topic. This was achieved by conducting an online survey, found in Chapter 2, aimed at parents of children with and without a congenital disorder. The second aim of this study was to determine and compare the diagnostic accuracies of two- dimensional facial analyses of congenital disorders. Face2Gene is a popular phenotyping web tool and is free to use for healthcare professionals. The technology does not, however, classify an individual as “non-syndromic” and will suggest likely syndromes to all submitted facial images. Differentiation between syndromic and non-syndromic individuals is important for clinicians to determine if the child requires further testing or investigation into a potential diagnosis. Chapter 3 aimed to establish how well Face2Gene can differentiate between syndromic and non-syndromic facial images, and we compared that to our in-house analyses of the facial features of individuals. Previous research showed that Face2Gene did not perform well in African ethnic groups before training. This is likely due to the algorithm’s training data mostly consisting of European individuals. It is also important to establish a diagnosis as early as possible, to ensure the correct management strategies are put in place. In Chapter 4, we thus aimed to establish how well the Face2Gene algorithm can differentiate between syndromic and non-syndromic facial images in different syndrome, ethnic, and age groups. We again compared that to the results from our in-house analyses.