A new artificial intelligence algorithm can detect whether a patient has a genetic condition using just a photo of their face, according to a study published in Nature Medicine and reported by STAT.
The study was conducted by researchers at FDNA, a Boston-based company attempting to use AI and facial analysis to help physicians diagnose genetic conditions. The researchers reviewed a collection of experiments testing how well an algorithm dubbed DeepGestalt detected genetic conditions compared to diagnoses from clinicians.
In one of the experiments, DeepGestalt was 64 percent accurate in identifying which of five genetic mutations might be causing Noonan syndrome, a genetic disorder that prevents development in various parts of the body. Yaron Gurovich, chief technology officer at FDNA, said the algorithm could help clinicians determine which potentially mutated genes to formally test in a lab.
"If you consider the phenotype properly, you are able to increase your odds of a diagnosis," he told STAT.
Mr. Gurovich added FDNA is working on another paper that goes beyond Noonan syndrome, demonstrating how the tool can be used more broadly. The study is currently undergoing peer review, but the pre-print version is available online on BioRxiv, an archive of unpublished science papers.
Still, other experts are skeptical of the invention.
Bruce Gelb, MD, director of the Mindich Child Health and Development Institute at New York City-based Mount Sinai, said this type of algorithm would not likely replace genetic tests currently in use to diagnose Noonan syndrome, particularly because Noonan syndrome has a variety of different genetic causes and symptoms. Dr. Gelb was not affiliated with the study.
"I don’t know why they undertook this, exactly," he said of FDNA's work in an interview with STAT. Although he acknowledged the algorithm's success rate was "impressive," he added he finds it "inconceivable to me that one wouldn’t send off the panel testing and figure out which one it actually is."
To access STAT's article, click here.