According to the researchers, algorithms could help detect facial features associated with rare conditions. They have explained how algorithms can support classify facial traits linked to genetic disorders, potentially accelerating clinical diagnoses.
In the report, US company FDNA published new experiments of their software, DeepGestalt. Alike other facial recognition software, the company trained their algorithms by assessing a dataset of faces. The company, using a smartphone app Face2Gene, amassed over 17,000 images covering 200 distinct syndromes. In their two first tests, FDNA’s DeepGestalt was used to look for particular disorders – Cornelia de Lange syndrome and Angelman syndrome, both are complex conditions that affect intellectual development and mobility. When tasked with differentiating between patients’ pictures with one syndrome or another, random syndrome, DeepGestalt was over 90 percent accurate, beating proficient clinicians, who were around 70 percent accurate on similar tests. When experimented on 502 images displaying individuals with 92 distinct syndromes, the software detected the target condition in its guess of 10 possible diagnoses over 90 percent of the time.
In a more challenging test, DeepGestalt displayed images of individuals with Noonan syndrome and calls to recognize which one of five specific genetic variations might have caused it. However, here the algorithm was slightly less accurate, with a hit rate of 64 percent, but it still showed much better than the 20 percent rate you’d get from guessing. Though, the specialists said that these kind of algorithmic tests aren’t a silver bullet for detecting rare genetic disorders. But, in the case of identifying specific genetic changes, Professor at the Icahn School of Medicine at Mount Sinai and an expert on Noonan syndrome, Dr. Bruce Gelb, stated in the statement that the definite answer from a genetic test would be more valuable.