A facial recognition scan could become part of a standard medical checkup in the not-too-distant future. Researchers have shown how algorithms can help identify facial characteristics linked to genetic disorders, potentially speeding up clinical diagnoses.
In a study published this month in the journal Nature Medicine, US company FDNA published new tests of their software, DeepGestalt. Just like regular facial recognition software, the company trained their algorithms by analyzing a dataset of faces. FDNA collected more than 17,000 images covering 200 different syndromes using a smartphone app it developed named Face2Gene.
In two first tests, DeepGestalt was used to look for specific disorders: Cornelia de Lange syndrome and Angelman syndrome. Both of these are complex conditions that affect intellectual development and mobility. They also have distinct facial traits, like arched eyebrows that meet in the middle for Cornelia de Lange syndrome, and unusually fair skin and hair for Angelman syndrome.
When tasked with distinguishing between pictures of patients with one syndrome or another, random syndrome, DeepGestalt was more than 90 percent accurate, beating expert clinicians, who were around 70 percent accurate on similar tests. When tested on 502 images showing individuals with 92 different syndromes, DeepGestalt identified the target condition in its guess of 10 possible diagnoses more than 90 percent of the time.
In a more challenging experiment, the algorithm was shown images of individuals with Noonan syndrome, and asked to identify which one of five specific genetic mutations might have caused it. Here the software was slightly less accurate, with a hit rate of 64 percent, but it still performed much better than the 20 percent rate you’d get from guessing.
However, experts say these sort of algorithmic tests aren’t a silver bullet for identifying rare genetic disorders. In the case of spotting specific genetic mutations, Dr. Bruce Gelb, professor at the Icahn School of Medicine at Mount Sinai and an expert on Noonan syndrome, told Stat News that the definite answer from a genetic test would be more useful.
“It’s inconceivable to me that one wouldn’t send off the panel testing and figure out which one it actually is,” said Gelb, who nevertheless said the algorithms were “impressive.”
Gelb also noted that DeepGestalt was developed and tested on a limited dataset of fairly young children, and might struggle to identify disorders in older individuals, where facial characteristics become less distinct. Third-party research of FDNA’s tools has also suggested a racial bias: the algorithms are much more effective on Caucasian than African faces.
FDNA seems aware of these shortcomings, and the company’s research refers to DeepGestalt’s potential as “a reference tool” — something that, like other AI-powered software, would assist, not replace, human diagnoses.
Christoffer Nellåker, an expert in the field at the University of Oxford, echoed this judgement, telling New Scientist: “The real value here is that for some of these ultra-rare diseases, the process of diagnosis can be many, many years […] For some diseases, it will cut down the time to diagnosis drastically. For others, it could perhaps add a means of finding other people with the disease and, in turn, help find new treatments or cures.”