By Allison Proffitt
September 25, 2018 | Can an artificial intelligence platform be trained to do expert phenotypic evaluation and suggest diagnoses for rare, genetic diseases based on a photo? FDNA believes so. Dekel Gelbman, the company’s CEO, said: “Everything that has to do with phenotyping should go through AI.”
The company is started with facial analysis. Its Face2Gene app—and underlying AI platform—is building a global library of facial images in an effort to link artificial intelligence, genomics, and precision medicine to improve rare disease diagnoses.
The Face2Gene app currently has images of 120,000 patients represented in its database, some with more than one photo. But the rate of growth since the app launched in 2014 has been accelerating. “We are expecting this to grow almost exponentially,” Gelbman told Diagnostics World. “We’re expecting to reach close to a million [patients represented] in a couple of years.”
FDNA has deep roots in facial analysis. The company was incorporated about seven years ago by Moti Shniberg and Lior Wolf, who had just sold their photo tagging technology to Facebook. As a next step, they wanted to introduce facial analysis to healthcare in a way that would have an immediate impact. Gelbman joined the team as CEO in January 2011.
“Very quickly, we stumbled into genetics,” Gelbman said. “This is something of a well-known secret in the industry—or a well-kept secret depending on where you come from—but geneticists actually look at facial patterns in their process of phenotypic evaluation and reaching a diagnosis,” he explained.
FDNA adds a layer of artificial intelligence to the process. The Face2Gene app is actually a platform hosting several cloud applications accessible through mobile apps and the web, explained Gelbman. The app is free to healthcare professionals and can be installed on iOS or Android devices.
Using the app is simple and quick. A doctor photographs a patient with the app, and the photo is uploaded to FDNA’s cloud-installed technology. The facial analysis software scans the photo and gives the user a ranked list of compatible diagnoses drawn from the company’s proprietary database. But it’s not a one-way street. The app walks physicians through questions to add additional, well-characterized, phenotypic data, further refining the results. Physicians close the loop by adding a final, molecular diagnosis if possible. It’s a virtuous cycle, Gelbman said.
“We benefit from the data that’s uploaded, and they benefit from the technology. So it’s a symbiotic relationship. We continue to provide technology to returning users… and they understand that if they give us feedback, the technology actually learns from that feedback and gets better,” he explained.
“There’s a very broad spectrum of users,” Gelbman said. “There’s a growing number of evangelists that have been early adopters and have been helping us with research… Eventually, what’s happened is that physicians who use this start to use it more and more frequently as their clinical warehouse or medical record for the patients in the genetic department.”
It’s not just physicians that benefit. Pathology labs, in particular, benefit from the platform’s structured phenotype data collection and the facial analysis. Pathologists with sequencing data analyze hundreds of genetic variants, often without ever seeing a patient, Gelbman said. They are missing the “single most critical piece of information.”
The AI solutions FDNA provides are more than just a pipe to get thorough, structured phenotype data to the lab so it can be matched with the genomic data. The AI can compare VCF files from a patient’s sequencing with facial analysis and rank the variants for likelihood of pathogenicity. “This information is extremely important for the labs,” Gelbman said.
If all this is free, what’s the business model?
The goal is to license the technology to research laboratories, pharmaceutical companies that focus on genetic disease, and other groups to be incorporated into their systems. “Imagine an EMR that can integrate this technology into their system and scan all of the patients,” Gelbman said. FDNA believes the technology could improve research and cohort analysis; it could improve clinical trial recruitment.
Licensed technology is the same technology FDNA is developing with clinicians, Gelbman emphasized, but an integrated solution is much more scalable. “The app is not scalable,” he said.
The Face2Gene platform is only as good as the data that are shared with it, and Gelbman said FDNA is actively working to expand the database. Disease presentation can change based on age, gender, and ethnicity. Already clinicians from over 130 countries are using the app, Gelbman said, though the US and Europe are the most heavily represented. FDNA is hoping to specifically expand representation through research projects. For instance, last year the company partnered with the Asia Pacific Society of Human Genetics to build a network of Asian physicians, and “help train the system with Asian ethnicities,” Gelbman said. “We’ve done similar projects in countries in Africa.”
Beyond diversity in the dataset, FDNA plans other expansions as well.
“When we talk to clinicians, they say, yes, facial patterns are a very strong indication for hundreds, maybe thousands, of diseases. But some patients come and they don’t have anything striking on the face,” Gelbman said. They may, however, have other telltale phenotypic signs on the hands, or in how they move, or in medical images, or even in voice patterns.
“Our goal as a company is to start expanding our phenotypic analysis to these other signals,” he said. It’s part of our vision to next-generation phenotyping. It’s not just a face.”