We may owe technology yet another vote of thanks for its contribution to disease and disorder research. A team of engineers and neuroscientists have developed a computer program capable of identifying, with extreme accuracy, the early emergence of rare genetic disorders through simple photographic portraits.
The study, published in the journal eLife, reveals that where genetic tests and the naked human eye fail to diagnose genetic disorders, carefully ordered strings of ones and zeroes may be of some assistance. The new program so far includes 90 disorders, many of which occur in fewer than six percent of the population. The disorders are so rare, doctors often can’t diagnose them using traditional methods, such as blood tests, because science hasn’t yet identified the genetic variants responsible for the conditions.
Even skilled professionals have a difficult time analyzing young children’s, says Alastair Kent, director of the charity Genetic Alliance UK, who was not involved with the latest study, according to New Scientist. "As a result,” he said, “families frequently experience long delays – years rather than months — before they receive a diagnosis for their child."
The software was developed by University of Oxford researchers Christoffer Nellåker and Andrew Zisserman. Together with a group of colleagues, the pair fed 1,363 pictures of people with genetic disorders (the photos were publicly available) into an algorithm designed to track for specific facial features of eight key disorders. Progeria, for example, is marked by a beaking of the nose and eyes that sink into their sockets. Relying on the bank of stored pictures, the software can read a sample image and scan it for known features. It then spits out a ranking of possible matches.
To demonstrate the software’s power, the team gave it a training test. They showed it pictures of people with eight known genetic disorders, for which the machine had between 100 and 283 images handy for reference. On average, the machine correctly predicted disorder 93 percent of the time. Now that it can screen for nearly 12 times as many disorders, with 2,754 faces at its disposal, Nellåker says it’s well on its way to helping pediatricians diagnose patients. "It's not sufficiently accurate to provide a rock-solid diagnosis, but it helps narrow down the possibilities," he said.
Photographic software is hardly new when applied to predicting future outcomes. In April, University of Washington researchers developed a piece of age progression software that can enhance the current ways people search for missing children. Like the University of Oxford’s program, the software culls thousands of pre-selected photos to form a composite prediction. These predictions were so accurate, in fact, the odds study participants chose the real image over the composite were virtually a flip of the coin.
Of the latest study, Kent said it was “a tremendous step forward in shortening the diagnostic journey that families embark on following birth of a child with dysmorphic features.” If validated in further tests of greater rigor, he continued, “it will provide access to expert advice and guidance for families more quickly and efficiently than is currently possible."
Source: Ferry Q, Steinberg J, Webber C, et al. Diagnostically relevant facial gestalt information from ordinary photos. eLife. 2014.