The University of Edinburgh
Challenges of human scored cleft aesthetic outcome evaluation: a critical review of literature and exploration of the potential for applying AI to analyse routinely collected global data
Abstract
dc:description.abstractBACKGROUND: Orofacial clefts are the commonest type of congenital anomaly to affect the craniofacial region, affecting approximately 1 in 700 live births in the UK, with higher incidences in low-middle income countries. Surgical repair of cleft lip and/or palate is usually performed in infancy; however, there are concerning inequalities in access to and quality of surgical care, and few global outcome studies. Scoring aesthetic results after surgery is crucial when determining the success of a repair, and helps to distinguish between advantageous and unfavourable approaches. Currently, a reliable and accurate scoring system utilising large numbers of unstandardised 2- dimentional photographs of patients from varying ethnic backgrounds, which is inexpensive, widely accepted and easily applicable, does not exist. Artificial Intelligence (AI) has been applied in a variety of specialities of medicine and surgery with beneficial results; however, its advantages have not yet been harnessed in cleft care. OBJECTIVES: This thesis had three primary objectives: 1. Toidentifytheproblemswiththepre-operativecleftseverityandpost- operative aesthetic scoring systems currently used in the management of cleft lips. 2. Inlightofthelimitationsofthecurrentmethods(specificallysurrounding human assessment), to consider the potential benefits and pitfalls of using AI in order to develop a robust reproducible scoring system. 3. ToevaluateanddescriberoutinelycollecteddatafromtheglobalSmile Train dataset (STX) and prepare the data for AI analysis. METHODS: 1. Toachievethefirstobjective,acriticalsurveyoftheliteratureconcerning pre-operative cleft severity grading and post-operative aesthetic outcome evaluation methods for cleft lip was undertaken. 2. AcriticalsurveyoftheliteraturerelevanttotheapplicationofAItosimilar challenges in medicine, in particular cleft and plastic surgery, was undertaken. The review focused on facial image analysis, the importance of ethical use, and minimising the risk of algorithmic bias. 3. Adatamanagementplanwascreatedforthetransferandstorageof demographic and image data from STX for over 1.5 million patients. Data were checked for accuracy and consistency. Demographic and diagnostic data were studied to determine inclusion and exclusion criteria. A dataset of images labelled with the diagnosis was created for AI analysis. CONCLUSIONS: A review of the literature demonstrated that current scoring methods lack sizeable datasets, ethnic diversity and applicable techniques in resource-limited centres, and there is need for an innovative globally applicable scoring system to evaluate aesthetic outcomes. Doing this will enable international comparative studies and audit/quality control with rapid closure of the audit loop, ultimately minimising inequalities and improving cleft care. AI can be used to analyse the dataset in preparation for future research when evaluating aesthetic outcomes. This study shows that, despite some challenges, it is possible to study routinely collected global data, thus enabling the application of AI to large datasets. Additionally, this study is able to make some recommendations to Smile Train for more robust data collection in the future.
Degree
thesis:*- Grantor dc:publisher
- The University of Edinburgh
- Year dc:date.issued
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Turnbull, Caitríona Elizabeth Anne
- Advisors dc:contributor.advisor
-
- Mehendale, Felicity Vidya
- Wigmore, Stephen
- Addison, Patrick
Subjects
dc:subject × 4Rights
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Identifier URI
- http://dx.doi.org/10.7488/era/5587