NEW YORK (360Dx) – Researchers at the Sheikh Zayed Institute for Pediatric Surgical Innovation have developed a digital facial analysis technology that they hope can assist doctors in diagnosing a variety of genetic diseases in diverse patient populations.
In a study published in the American Journal of Medical Genetics last month, a team led by the National Human Genome Research Institute showed that it could use the software to accurately detect Down syndrome in patients with diverse ethnic backgrounds.
"This is a way to augment the clinical identification of a genetic condition, and a tool that can help the clinician to make better and more objective decisions when they see their patients," said Marius George Linguraru, a principal investigator at the Washington-based Sheikh Zayed Institute at Children's National Health System who developed the facial analysis software, and an author of the study.
He and his colleagues are currently testing the technology on patients with other genetic diseases and plan to conduct clinical validation studies, with the goal to take the tool through regulatory approval and commercialize it in order to make it available to doctors around the world.
Results from the published study are going into the National Institutes of Health's Atlas of Human Malformation Syndromes in Diverse Populations, a freely available resource that was launched last April.
The atlas helps doctors diagnose patients of diverse backgrounds by comparing their phenotypes and symptoms with photos and descriptions of patients of the same ancestry who have confirmed diagnoses. The first two conditions covered by the atlas were Down syndrome and 22q11.2 deletion syndrome, also called DiGeorge syndrome or velocardiofacial syndrome, and it continues to grow. When completed, the atlas will contain data from patients with a variety of conditions from the Middle East, the Indian subcontinent, other regions in Asia, South America, and sub-Saharan Africa. It will be searchable by phenotype, syndrome, geographic region, and genomic or molecular diagnosis.
NIH decided to create the atlas after realizing that inherited diseases can manifest differently in different ethnic groups and require reference photos from specific populations. However, photos currently used for diagnostics tend to come from patients with northern European ancestry, who often have different disease characteristics than patients from other regions of the world. For example, in Down syndrome, a characteristic feature of patients with European ancestry is a skin fold of the upper eyelid that covers the inner corner of the eye. However, such eye folds are normal in healthy people of Asian descent.
Also, doctors find it difficult to diagnose Down syndrome clinically at birth, with one study suggesting that clinical diagnoses of newborns are accurate only 64 percent of the time. In addition, prenatal screening for congenital conditions is still uncommon in many parts of the world, although the advent of noninvasive prenatal testing might change that.
The team's digital facial analysis approach might put "very simple tools into the hands of clinicians who may not have access to state-of-the-art equipment or genetic testing in their local clinics," Linguraru said.
An early diagnosis can make a difference for the patient, he added. "The earlier you can identify that a child has Down syndrome, the better we can think about the management of that individual," he said, including early treatment of cardiac conditions.
His team joined the NHGRI research consortium "to bring more objectivity to the study" with its software, which can precisely quantify facial features of Down syndrome patients.
The software analyzes a frontal facial photograph of the patient, using a combination of machine learning and quantitative imaging. The technology is similar to facial recognition software employed at airports or by companies like Facebook and Google, but the researchers developed it for medical applications, so it can identify patterns that are characteristic of a disease.
For their published study, the researchers evaluated 65 Down syndrome patients from 13 countries, ranging in age from one month to 26 years, whose diagnosis had been confirmed by cytogenetic testing. They also analyzed 64 cases and 132 healthy ethnically matched controls from previous studies. Their algorithms extracted 126 facial features from a set of 44 facial landmarks and selected the most significant ones.
Overall, the software was able to detect Down syndrome with a sensitivity of 96.1 percent, specificity of 92.4 percent, and accuracy of 94.3 percent. Accuracy improved when the technology was applied to distinct population groups — 95.6 percent for Caucasian, 97.9 percent for African and African American, and 100 percent for Asian patients.
Meanwhile, the researchers have conducted a similar study on DiGeorge syndrome, which they plan to publish in the next few months, and are planning another study on Noonan syndrome. "For each of these studies, we hope to have a similar perspective in which we identify what is similar and what is different between these populations and provide a better reference for identification of these syndromes in ethnically diverse groups of individuals," Linguraru explained.
The software could potentially be applied to all inherited conditions that come with facial dysmorphology, often conditions that involve the brain. "The face, in a way, is a reflection of the brain — the development of the brain is mirrored in the development of the face," he said.
Eventually, the researchers hope to develop their technology into a tool for doctors, who would take a picture of their patient on their cell phone and have it analyzed by the software. "This is a research product at this stage, but our plan and hope is to find a pathway to commercialize it because a commercial product is the way to make this available all around the world," Linguraru said.
Turning the software into a clinical tool will require clinical validation studies and fulfilling regulatory requirements by the US Food and Drug Administration, he said. The researchers are already collaborating with US-based and international clinical institutions to conduct such studies "as fast as possible" and are looking for additional clinical partners, he added.
The Sheikh Zayed Institute's group is not the only one that is trying to harness digital facial analysis for inherited disease diagnostics. FDNA, for example, a Boston-based startup that was founded by the developers of Facebook's facial recognition technology, has also created software that analyzes photos of patients with inherited diseases and matches them with specific syndromes. Last fall, the company presented an updated version of its Face2Gene software suite at the American Society of Human Genetics annual meeting.