1182 Ethnic Diversity in Facial Image Databases – the State of Play for Aesthetic Plastic Surgery

Abstract Aim The rise of social media has resulted in heightened patient expectations of cosmetic outcomes. The increasing trend of selfies and image collection may promote higher-powered predictive analytics to inform research and pre-operative consultations supporting informed consent. However, given the scattered and siloed nature of such databases, little is known regarding their composition. This study evaluated published facial image databases for ethnic representation, image quality and labelling. Method Databases were identified via a systematic search on PubMed, Embase, PsychInfo and Google Scholar. Databases were evaluated by two independent reviewers for demographics alongside image characteristics and governance surrounding data availability and usage. Results 6,544,444 images of 199,579 people from 54 databases were included in the study from 1988 – 2023. The USA (n=19, 33%) and UK (n=1, 18.2%) accounted for more than half of the databases. Gender distribution across the 20 databases that provided information on this showed an approximately equal split, with 216,331 images of males and 216,459 images of females. Of the data labelled for ethnicity, the most common was Black (40.1%, n=48,425) followed by White (38.7%, n=46,738) and Asian (8.3%, n=10,061). Only one database (FERET) had longitudinal coverage of age progression. 59% of databases were publicly available, whilst 87% did not permit commercial use. Conclusions Labelling of large datasets is limited without important information on ethnicity and gender. More work needs to be done to ensure that these databases are suitable for more widespread research purposes and to promote greater representation within such databases and aesthetic plastic surgery research.