Researchers Use Face Biometrics to Identify Genetic Disorders

Researchers from FDNA and the University of Bonn are trying to use facial biometrics to improve their chances of diagnosing people with rare genetic disorders. FDNA is the developer of the Face2Gene app, though it recently replaced its original algorithm with a new one that was developed in collaboration with the University.

While Face2Gene does rely on facial biometrics, the app itself is not designed for identification. It instead tries to group the faces of people with the same condition, so it can learn to recognize common traits that could lead to a diagnosis in another case.

As Wired reports, the research is predicated on the idea that many genetic disorders correlate with distinct facial characteristics. The problem is that while some genetic disorders (such as Down syndrome) are relatively common, others are rare to the point that most doctors have never encountered them in their practice, and therefore do not have the knowledge to address them properly.

The same was true of the original Face2Gene app, which could spot about 300 of the most common disorders with a high degree of accuracy. However, FDNA needed at least seven photos to train the algorithm to read a condition, and many of the rarest disorders to not have enough documentation to reach even that low threshold.

The new GestaltMatcher algorithm was introduced to Face2Gene last month, and while it is slightly less accurate when applied to the 300 most common conditions, it performs better across a much wider range of disorders. GestaltMatcher can classify roughly 1,000 different conditions, even when there are as few as two known cases. In doing so, it can help doctors rule out other, similar conditions, and allow them to reach a more accurate diagnosis.

A successful classification does not dictate a specific treatment path, though it could inform future research on a particular subject. The researchers were given the chance to test the new algorithm when two unrelated children in different countries developed the same condition in 2017, leading to a report that was published in 2019.

“That was kind of the first time that it worked,” said FDNA Chief Scientific Officer Peter Krawitz, who is also the head of the genomics institute at the University of Bonn. “We’re able to now work on disorders that the system didn’t learn or wasn’t trained on.”

Researchers at Okayama University have previously used facial recognition to study the effects of Parkinson’s disease. The GestaltMatcher algorithm could be a valuable diagnostic tool (especially when used in conjunction with other tests), though it is worth noting that small sample sizes have created bias and accuracy issues in facial recognition systems.

Source: Wired

–

March 9, 2022 – by Eric Weiss