People with rare genetic syndromes sometimes have indicative traits on their faces. Now for the first time an artificial intelligence system, developed by researchers in the US, can read such a genetic condition by simply analyzing photographs of individuals.
Researchers of Boston’s FDNA Biotechnology Company, headed by Yaron Gurovic, who published a publication in the journal Nature Medicine, according to Science and New Scientist, developed the DeepGestalt neural network, which analyzes the and draws a list of the ten most likely genetic syndromes a person may have.
The system, using computational vision and special deep-learning algorithms, was trained with 17,000 photos corresponding to over 200 genetic syndromes. DeepGestalt was then tested in another 502 photos and gave the right answer (included the right condition in the top ten) in 91% of cases.
In the next step, the system was able to distinguish between different genetic mutations that lead to the same syndrome. Analyzing human imaging with Nunnan syndrome, which inhibits physical development and can be derived from five different gene mutations, the system managed to detect the correct genetic mutation in nearly two-thirds of cases (64%).
“It’s definitely not perfect, but it’s nevertheless much better than what people can do,” Gurovic said. DeepGestalt proved to be superior to doctors in the diagnosis of the rare syndromes’ Angelsmann and Cornelia de Lange.
System makers have pointed out that DeepGestalt will help physicians and patients considerably, making timely, inexpensive and accurate diagnoses. But the fact that the diagnosis is made only by a simple photograph raises issues of privacy protection.
In the future, employers, insurance companies and other stakeholders will be able to know which genetic conditions a person has by simply analyzing his photos, which could lead to discrimination against him.