Skip to main content
Premium Trial:

Request an Annual Quote

FDNA Bets on Facial Recognition Software to Boost Diagnosis of Rare Genetic Syndromes


NEW YORK (GenomeWeb) – Many genetic syndromes come with characteristic facial traits and malformations, and doctors have been considering these for making diagnoses. Using facial recognition software, startup FDNA has developed a technology that allows this process to be automated and combined with phenotypic and genetic information.

By integrating facial analyses with other phenotypic traits and variants from genetic tests, such as exome and genome sequencing, the company hopes to boost the diagnostic rate among patients with rare and difficult-to-classify genetic disorders.

"Over the past year, it became more apparent that phenotyping is essential for that process — genotyping and phenotyping really walk hand in hand and are part of the same solution," Dekel Gelbman, FDNA's CEO, told GenomeWeb. "What we aspire to be is the leading phenotyping technology for this industry. And I think that through our collaborations, not only with clinicians but with all the diagnostic labs we've recently started collaborating with, we're poised to be that technology in the very near future."

Earlier this month at the American Society of Human Genetics annual meeting, FDNA launched an updated version of its Face2Gene software suite, which consists of five applications. The platform, which is geared at clinical geneticists, is free of charge for healthcare professionals and has approximately 2,500 users around the world.

Using the Clinic app, doctors can upload a facial photograph of their patient. The FDNA software then analyzes the photo and extracts a set of de-identified biomarkers, which it compares to a database of patients with confirmed genetic diagnoses. Based on facial similarity, it then generates a list of possible syndromes and associated phenotypic traits. Doctors can use this information to further evaluate their patient, while laboratories, through the Labs app, can use it to filter and prioritize variants from genetic tests.

In addition, through the firm's Library app, users can access London Medical Databases, a database of genetic syndromes, for a subscription fee. Meanwhile, the Forums app provides a platform for doctors to collaboratively review difficult-to-diagnose cases, and the Academy app is an interactive dysmorphology training tool.

FDNA was started almost six years ago by the founders of, a company that had developed a facial recognition technology and was later acquired by Facebook. FDNA, which is located in Boston and has an R&D center in Israel, has since grown to approximately 40 employees and is privately funded. Most of its investors are former or current executives from the pharmaceutical industry, Gelbman said.

When the firm started out, he recalled, the founding team was looking for an application of its technology in healthcare and soon came across clinical genetics, in particular facial dismorphology. An estimated 50 to 60 percent of rare genetic diseases manifest in facial malformations, he said, and between 2,500 and 4,000 genetic syndromes are either characterized by distinct facial patterns or associated with specific malformations.

Over the next three years, the company developed its Facial Dysmorphology Novel Analysis, or FDNA, technology, which required training the system with data from established cases and "trying to teach the computer to see things as a dysmorphologist would," Gelbman said.

The result was Face2Gene, the company's first product, which launched in March of 2014. In its first incarnation, the software was more of a "search and reference tool" that allowed clinicians to search an image against a database of syndromes with associated images, he explained, a resource that has grown to hundreds of thousands of patients. The system complies with privacy and HIPPA rules, he added, and photos are stored securely and only used initially to extract de-identified markers.

Over the past two years, FDNA has developed the technology further, employing a supervised learning system that enables it to find patterns that have never been described in the literature before. This, he said, "has allowed us to extend way beyond those known patterns of human malformations into unchartered territory," and to establish correlations between facial malformations and other phenotypic traits. With the help of the users of its software, the company is now ready to "get into deep phenotyping of syndromes and patients."

This works through the back-and-forth interactions between users and the software, he explained. Clinicians uploading a patient photo first obtain a list of suggested syndromes, along with a list of potential phenotypic traits. They then select which traits their patient actually has or doesn't have, which prompts the computer to make additional suggestions, and "through very few iterations with the clinician, we get to very detailed lists of phenotypic traits of a patient," Gelbman said.

This method, he said, "allows us to crowdsource hundreds of clinical geneticists' evaluation processes, which then goes to teach the system how to get better and better with time. Every time a clinician uses our system, our system learns from the clinician and gets better for the next time."

To entice clinicians to use Face2Gene more often, FDNA has put up a "60-day challenge" that awards users who upload a certain number of cases within 60 days with free exome sequencing for undiagnosed cases — provided by sponsoring laboratories GeneDx, Blueprint Genetics, MNG Laboratories, and Prevention Genetics — as well as a free subscription to London Medical Databases.

FDNA also has an ongoing research program, called "Give a Face to a Syndrome," to improve the system further. Under this program, the company collaborates with researchers to identify, for example, new facial patterns for known syndromes, or for specific genetic mutations causing a single syndrome. Such facial patterns might be too subtle for clinicians to observe but computers might be able to pick them up, Gelbman said.

Some scientists, for example, want to show that autism, which has not traditionally been associated with facial characteristics, has detectable facial patterns, while others have been working on patterns associated with fetal alcohol syndrome. The company's research is also branching out into disease areas beyond clinical genetics now, he added, including neurology and psychiatric disorders, and applications such as discovering early signs of stroke.

The next step, Gelbman said, is to integrate the phenotypic information with genetic test results in order to help with the interpretation of genetic variants. To that end, FDNA has teamed up with researchers at the Charité hospital in Berlin, Germany, for the Prioritization of Exome Data by Image Analysis (PEDIA) study.

For this study, which has a number of collaborating laboratories and clinicians and is open to additional ones, labs submit facial photographs of syndromic patients to FDNA, along with their confirmed molecular diagnosis and genetic variant file. Earlier this year at the European Society of Human Genetics annual meeting, a Charité researcher presented results from his team's first performance test of the software and talked about its plans for using it to prioritize exome variants.

Aside from the research study, FDNA now also allows genetic laboratories, through its Labs app, to access a phenotypic report for a patient from the submitting clinician and to review a list of suggested syndromes and clinically relevant genes, and Gelbman said that more than 20 laboratories have started using this tool now.

However, one challenge for labs right now is that physicians rarely submit a photograph of their patient at the moment, said Sherry Bale, a co-founder of genetic testing company GeneDx. "So in clinical testing of exomes, it would not be useful. Unless of course we could change the way physicians practice."

The hope is that clinicians will submit genetic variants they receive from the lab to Face2Gene, further improving FDNA's database. "That goes back and trains our system and opens up the door for the future to not only associate facial images with syndromes or traits but actually go down to the molecular level and associate specific patterns in the face with specific variants and mutations," Gelbman said.

Improving genetic variant analysis is likely going to be an important area for FDNA going forward. "We think that's going to be huge," Gelbman said. "Our vision is that our technology is going to be a standard phenotyping technology that's coupled with an exome or whole-genome sequence [test] in the future."

But while FDNA's software is intended to help physicians and laboratories make better diagnoses, he stressed that it is not designed to be a diagnostic. "It is a clinical and genetic resource that augments the clinical evaluation process and helps the clinician reach a diagnosis. But in itself, it's not a diagnostic tool, nor do we intend for it to be a diagnostic tool," Gelbman said.

Instead, by having clinical users help the company improve its phenotyping technology over time, FDNA hopes to achieve a level of sophistication that allows it to apply it to paid pharmaceutical research collaborations, in particular with companies developing orphan drugs.

Gelbman said FDNA already has several ongoing collaborations, though he declined to disclose the partners, and that its technology could be used for clinical trial management and screening, for the discovery of biomarkers for companion diagnostics, and even for targeted drug discovery.

"We're seeing the orphan drug industry grow at a tremendous pace. Over the last 10 years, the number of indications for rare diseases has grown really exponentially," he explained. "There is a lot of interest from the industry, and some collaborations are underway with some of the leading pharma companies in this field."