article · Clinical Genetics
Clinical assessment of facial dysmorphism assists in diagnosing genetic conditions, but interpretation can vary by ethnic background. An evaluation of 127 African children with intellectual disability showed only fair agreement between five African and five European clinicians, yielding a kappa coefficient of 0.29. In parallel, an automated facial recognition platform, FDNA Face2Gene, initially performed substantially better on Caucasian faces (80 percent recognition) than on African faces (36.8 percent recognition) for Down syndrome. Following training with a dedicated set of African photographs of patients with and without Down syndrome, the recognition rate in African patients rose to 94.7 percent. These findings indicate that human clinical assessments depend on evaluator background, while computerised tools depend heavily on patient ethnicity and benefit significantly from regionally representative training datasets.
Accurate identification of facial features is crucial for diagnosing developmental conditions, but diagnostic tools and clinical judgment can both suffer from ethnic bias. Demonstrating that automated diagnostic tools can be retrained to substantially improve detection in underrepresented groups shows a practical route toward fairer and more reliable clinical genetics assessments across diverse global populations.
This research applies to automated clinical diagnostic software used by healthcare providers evaluating genetic disorders such as Down syndrome. The tested technology, Face2Gene, is an applied digital health tool. The findings show that incorporating ethnically specific training datasets into such platforms can directly improve detection sensitivity, offering commercial developers a pathway to adapt existing diagnostic solutions for broader international markets.
AI-generated from the published abstract. Always read the original work before citing.
The evaluation of facial dysmorphism is a critical step toward reaching a diagnostic. The aim of the present study was to evaluate the ability to interpret facial morphology in African children with intellectual disability (ID). First, 10 experienced clinicians (five from Africa and five from Europe) rated gestalt in 127 African non-Down Syndrome (non-DS) patients using either the score 2 for 'clearly dysmorphic', 0 for 'clearly non dysmorphic' or 1 for 'uncertain'. The inter-rater agreement was determined using kappa coefficient. There was only fair agreement between African and European raters (kappa-coefficient = 0.29). Second, we applied the FDNA Face2Gene solution to assess Down Syndrome (DS) faces. Initially, Face2Gene showed a better recognition rate for DS in Caucasian (80%) compared to African (36.8%). We trained the Face2Gene with a set of African DS and non-DS photographs. Interestingly, the recognition in African increased to 94.7%. Thus, training improved the sensitivity of Face2Gene. Our data suggest that human based evaluation is influenced by ethnic background of the evaluator. In addition, computer based evaluation indicates that the ethnic of the patient also influences the evaluation and that training may increase the detection specificity for a particular ethnic.
This page summarises published work. The authoritative version sits with the publisher.
DOI: 10.1111/cge.12948
Is something wrong with this record? Report it or request removal.
Discussion
Have you built on this work, tried to replicate it, or seen it applied in practice? Share what you know. Verified researchers and MARATTO™ domain experts can open a discussion, and any member can reply. Contributions are reviewed before they appear.
No discussion yet. Open the first thread.
New to MARATTO™? Create a free account.
