A new artificial intelligence technology can accurately diagnose some genetic disorders using only the patient's photograph.
The technology called DeepGestalt was developed by Yaron Gurovich and his team at FDNA, a biotechnology firm in Boston.
The research team built a neural network to look at gestalt, or overall impression of faces, and return a list of the genetic syndromes that a person is most likely to have. The AI outperformed human clinicians in identifying the genetic syndromes in three trials.
People with genetic syndromes usually have revealing facial features and some disorders often show up in their appearance. Individuals with Noonan syndrome, for example, can have wide-set eyes. It is a genetic condition that inhibits the body's growth and development. Those with brain-type intellectual disability, which is caused by a mutated gene on the X chromosome, have almond-shaped eyes and small chins.
Now, with DeepGestalt computer program, researchers have trained AI to recognize these facial features, paving the way for early and cheap diagnoses. The "deep learning" program used more than 17,000 photos of patients with more than 200 rare genetic disorders to recognize indicators that point to different syndromes.
"It demonstrates how one can successfully apply state of the art algorithms, such as deep learning, to a challenging field where the available data is small, unbalanced in terms of available patients per condition, and where the need to support a large number of conditions is great," said Yaron Gurovich, the lead author of the study.
Among 502 chosen images, the team found that the AI identified the correct syndrome among its top 10 list 91 percent of the time.
Privacy Breach And Other Issues
The researchers admit that one of the challenges this new technology is facing are issues of privacy breach, which could be abused by employers or insurance providers. The distribution and use of the tool should be regulated and it should only be available to clinicians.
Another difficulty is that the performance of an Artificial Intelligence system is difficult to evaluate. Gurovich said that the reason why it is hard is that there are not sufficient publicly available benchmarks.
Despite these difficulties, experts are hopeful that this new technology could bring benefits to people with genetic syndromes.
The study was published on Monday, Jan. 7 in the journal Nature Medicine.