Facial analysis technology aids diagnoses of genetic disorders

Facial analysis technology that is becoming easier to use and more accessible may help geneticists narrow down possible genetic diagnoses behind disorders that often involve dysmorphic facial features. Researchers at the University of Oxford in the United Kingdom have developed a computer application based on an algorithm that employs facial analysis technology to extract various phenotypic information about facial dysmorphisms from nonclinical photographs. The application uses machine learning, a type of artificial intelligence that learns from data instead of following explicit programmed instructions. It then builds a description of the face structure, known as the facial mesh, and compares it against data from other facial meshes in the system before delivering a list of possible genetic diagnoses. The algorithm becomes better at spotting facial phenotypes associated with a disorder as it analyzes more photos of faces with those specific features [Ferry et al., 2014]. In contrast to older facial analysis systems that rely on costly equipment to analyze three-dimensional (3-D) images, the Oxford-based researchers used easily accessible two-dimensional (2-D) photographs of children's faces to analyze specific facial features that are associated with up to 40% of genetic disorders. Project leader Christoffer Nellåker, PhD, co–senior author Andrew Zisserman, PhD, and colleagues used their system to analyze craniofacial features in 2,754 photographs of patients diagnosed with 90 known syndromes and found that the technology aided geneticists in determining the correct diagnosis by 27.6-fold. The disorders included progeria as well as Angelman, Apert, Cornelia de Lange, Down, fragile X, Treacher Collins, and Williams-Beuren syndromes. Using additional reference images, the algorithm correctly predicted the disorders 93% of the time on average, the researchers write. The researchers then expanded their analysis to photos of children with 82 other disorders, including various mutations in PACS1, specific genes in the RAS/MEK pathway, 22q11 deletion, and Marfan and Sotos syndromes, and determined their approach also eased diagnosis in these cases by 27.6-fold. The researchers' application may help a geneticist decide to test for a more common disorder, but it could also help identify an ultra-rare disorder, says Dr. Nellåker, Research Fellow in the Medical Research Council Functional Genomics Unit in the Department of Physiology, Anatomy, and Genetics at the University of Oxford. “There are many different types of rare disease, including many you may see once in your career,” Dr. Nellåker says, adding that their approach might identify ultra-rare diseases if it has meshes from multiple individuals with similar but very rare facial phenotypes. Although 3-D facial analysis can also aid in the identification of rare diseases through much more highly detailed images and analysis, the equipment necessary for 3-D analysis is expensive and requires images dependent on children sitting still and maintaining consistent facial expressions. In contrast, the researchers' 2-D system is designed for widespread use and requires a camera or ordinary photographs, a scanner, and a computer, he adds. Another facial analysis program known as Face2Gene, launched recently by New York–based FDNA Accessible Genetics, is a genetic search and reference mobile application that has capabilities similar to the approach developed by the Oxford researchers. It can be downloaded for free to Apple devices. Face2Gene is the result of several years of research, development, and validation by a group of geneticists led by Michael R. Hayden, MD, PhD, Chairman of FDNA's Scientific Advisory Board and Steering Committee and Senior Scientist at the Centre for Molecular Medicine and Therapeutics in Vancouver, Canada. The application, based on processing tens of thousands of facial images, creates and uses facial meshes to identify particular facial phenotypes through learning algorithms and a crowd- sourcing method. “The more clinicians upload cases into Face2Gene, the better the technology becomes for everyone in the network,” says Dekel Gelbman, LL.B, MBA, Chief Executive Officer of FDNA. Face2Gene refers users to both genetic databases and other clinicians whose patients have similar facial meshes and can share information about particular cases within the wider network of users. “If there's a case you're struggling with, you can upload images into the application [and] others can comment if they have seen something similar,” Mr. Gelbman says. This sort of communication previously occurred mostly at annual conferences, he notes. Users are enthusiastic about Face2Gene. “It gives confidence that you are ordering the right test,” says Karen W. Gripp, MD, Division Chief, Genetics Director, and Costello Program Chief in the Division of Medical Genetics at Alfred I. duPont Hospital for Children in Wilmington, Delaware. The feature that connects users to genetic databases is useful to “see if I'm missing something,” she adds. Omar Abdul-Rahman, MD, Professor of Pediatrics and Neurology at the University of Mississippi Medical Center in Jackson, says he uses Face2Gene as a reference when he suspects a genetic condition. He says the application provides a list of possible conditions, which is useful when a child lacks distinctive facial features but has other key features of a particular disease. Dr. Abdul-Rahman says the application was recently useful in his decision to test, and ultimately diagnose, a child with Mowat-Wilson syndrome [see “Importance of Facial Features in Mowat-Wilson Syndrome Highlighted,” p. X], which he had suspected but previously had never seen in practice. He has also used Face2Gene to help identify a spectrum of diagnoses associated with alcohol exposure. “You have to recognize that facial analysis technology is still a new tool and that it has limitations,” Dr. Abdul- Rahman emphasizes. “It is useful in the hands of a geneticist who still must rely on clinical judgment.” “Absolutely, these types of technology are tools to aid the work of clinician experts and to narrow the search space for possible diagnoses,” agrees Dr. Nellåker. “Both our and the FDNA approach are machine-learning methods and with more images will become better at accounting for ethnic, gender, and age variation.”

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.