Spotting rare genetic disease is, by definition, a little tricky. But now a team of scientists from the University of Oxford has developed software that can spot such conditions by analyzing family photographs.
Many genetic conditions are known to exist but impossible to test for, because the gene variants that cause them haven't actually yet been identified. Instead, doctors often rely on pronounced facial features to make a diagnosis, because up to 40 percent of rare disorders give rise to distinctive differences in appearance.
If that sounds like it lacks rigor, it's because it does; very few clinicians are trained in this way, and even then it's hard work and open to subjectivity. Step in Christoffer Nellåker and Andrew Zisserman, who work in the field of computer vision and have developed a machine learning algorithm that is able to recognize a variety of rare genetic conditions from simple digital photographs, reports New Scientist.
Their algorithms were initially fed 1,363 publicly available pictures of people with eight genetic disorders—including Down's syndrome, fragile X syndrome and progeria—and spotted 36 distinct facial features that could be used to distinguish between the conditions. It worked well, but it's been expanded to spot over 90 disorders, and it's claimed that it makes it 30 times more likely that a patient will be correctly diagnosed. The results are published in eLife.
It's impressive stuff, and will be most valuable in countries where medical resources are scant. In developing countries where clinical genetics is missing entirely, children may soon be diagnosed with more accuracy than they could ever have hoped. [eLife via New Scientist]