Better Than DNA Sequencing?

A new computer program known as DeepGestalt was recently developed to identify and quantify hundreds of genetic syndromes using facial analysis. The artificial intelligence (AI) system was “trained” using a public database comprising more than 17,000 images of patients with more than 200 rare genetic disorders. By breaking the images down into mathematical facial descriptors and comparing them to patterns of markers for known genetic syndromes, the program can generate a prioritized list of diagnoses.

The system was tested with over 500 images of individuals with Noonan syndrome (a genetic disorder that is characterized in part by wide-set eyes) or Bain type intellectual disability (which is sometimes indicated by almond-shaped eyes and a small chin). The program included the correct diagnosis in a list of top ten diagnoses 91 percent of the time. It was also better than doctors at identifying subtypes of Noonan syndrome and spotting other inherited mutations that are revealed by characteristics such as low-set ears and an upturned nose. In comparison, it has been estimated that the odds of identifying specific genetic disorders through DNA sequencing has been estimated to be about 25 percent because the relevant variants are not currently well defined.

When used in combination with genome testing, this system could help physicians more accurately diagnose genetic disorders while reducing the time and cost. However, such a tool will need to be strictly regulated to prevent exploitation by employers, insurance providers and other entities.