ESPE Yearbook of Paediatric Endocrinology

In Brief: The authors tested the diagnostic ability of the Face2Gene app in 23 children with a clinical or genetically confirmed diagnosis of Silver-Russell syndrome (SRS) and 29 children with genetically confirmed Prader-Willi syndrome (PWS). When combined with careful routine clinical history and examination, the Face2Gene app can be a useful diagnostic tool.

Among PWS patients, Face2Gene calculated the top 1, top 5, and top 10 sensitivities to be 76%, 97%, and 100%, respectively. ‘Top 1/5/10’ means that the correct syndrome is listed among the top 1/5/10 possible syndromes. Face2Gene app performance was slightly better in PWS patients with paternal deletion of chromosome 15q11-13 than those with maternal uniparental disomy, but showed no difference with age.

Detection of SRS was overall lower: top 1, top 5, and top 10 sensitivities were 39%, 65%, and 91%, respectively – but Face2Gene performed better in younger SRS patients, or if a photo of the child taken at a younger age was available.

Clinical geneticists (and many of our esteemed colleagues) are particularly good at spotting rare genetic disorders based on a patient’s facial and other physical characteristics. But genetic testing is becoming more widely available - without needing referral to clinical genetics clinics. This raises the risks that appropriate genetic testing may be omitted or that expensive tests are too frequently requested. Artificial intelligence is extremely good at pattern recognition. The Face2Gene app takes a single frontal photo and uses face detection technology built on deep convolutional neural networks. Specific facial landmarks are extracted and compared to the app’s database of information on >10 000 syndromes.

There will be concerns about safety and confidentiality. Face2Gene seems to be secure – original photos uploaded to it are encrypted and stored securely, available only to the individual clinician or researcher who submitted the case - although this will need to be tested and validated by regulators. However, these are striking new findings and likely represent a taste of the many benefits to come for clinical practice in the world of artificial intelligence.