Computers have gotten creepily good at recognizing faces. But a program developed by two researchers does more than that: it looks at photos of people's faces and helps diagnose genetic diseases or disorders. The two researchers, Christoffer Nellåker and Andrew Zisserman of the University of Oxford, have taught the program to identify a host of genetics disorders, including Down's syndrome and fragile X. When it analyzed Abraham Lincoln's face, it suggested that he may have had Marfan's syndrome—a disorder that some think affected the 16th president.
Nellåker and Zisserman developed their program by feeding it nearly 1,400 photos of people with eight different genetic disorders. They then taught it recognize characteristic features of each. In tests, it accurately identified these disorders in photographs of new faces 93 percent of the time, on average.
Perhaps surprisingly, between 30 and 40 percent of genetic disorders affect the shape of the face or skull, the authors write in the study. And it's these signatures the software is designed to pick up. The scientists have improved the program since describing it in a study now published in the journal eLife, as New Scientist reported:
The team have since expanded the software so that it recognizes 90 disorders. It can't give an exact diagnosis yet, but based on the 2754 faces now in the database, the researchers estimate that the system makes it almost 30 times more likely that someone will make a correct diagnosis than by chance alone.
"It's not sufficiently accurate to provide a rock-solid diagnosis, but it helps narrow down the possibilities," Nellåker added. But this is far from conclusive. It's not enough to prove, for instance, that Lincoln did have Marfan's syndrone. (Not that humans are sure either: one researcher, for instance, thinks Lincoln had a rare, cancer-causing genetic disorder called MEN2B.)
However, the program could be useful in allowing parents to decide if they want to see a doctor or a geneticist for further testing, or to narrow down potential cases for clinicians to evaluate.