Down Syndrome Research Has Been Biased. Will Artificial Intelligence Help?
South African researchers are developing facial scanning technologies to help diagnose Down syndrome earlier in black children.
Ten cameras are pointed at 4-year-old Karabo.
Karabo, though, is looking at the ceiling, her attention turning to the harness strapping her into the baby seat, anywhere except the cameras. Meanwhile her mother, a professor, and a graduate student are all struggling to make her look at even one of the cameras so they can capture her 3-D image.
Karabo is one of many children who are part of an ongoing study of Down syndrome in African populations at the University of Pretoria, where scientists are making a database of 3-D facial images and plan to develop an algorithm to recognize the distinctive facial features that could potentially point to genetic conditions based on facial imagery. That project coincides with another South African project aiming to use artificial intelligence to facially scan those who might be suffering from a disease but not know it.
Down syndrome, a genetic condition in which the individual has a duplicate of chromosome 21, is underreported among the black population in South Africa and globally. As a seminal 1997 study published in the South African Medical Journal pointed out, Down syndrome is significantly under-reported in black children because under-trained public hospital staff—where black children are disproportionately born—do not recognize the physical markers of the disease.
Babies with Down syndrome often have distinctive facial features, such as flatter noses and extra folds of skin at the corners of their eyes. But those physical attributes aren’t the same across the board, and research is lacking in the identification and treatment of Down syndrome for black children.
“Very little research is done on African populations, and the facial features actually look quite different in African populations compared to European populations,” Vinet Coetzee, a geneticist at the University of Pretoria, told The Daily Beast. Coetzee, who heads up the research program, aims to have the database to visually diagnose Down syndrome completed by the end of next year, then plans to expand the research to about 700 other conditions that have facial markers over the next decade.
“These features don’t get recognized, and it’s the same with other conditions—something that is obvious in one population might not be in others,” Coetzee pointed out.
In the United States, Down syndrome is the most common chromosomal disorder, affecting about one in every 700 newborns, according to the Centers for Disease Control and Prevention. However, corresponding data is not available for most African countries. Another study in South Africa’s coastal KwaZulu-Natal province published last year found that clinicians were only likely to correctly identify Down syndrome in newborns half of the time, compared to 83 percent in the Netherlands.
“The main thing we want to change is that many children only get diagnosed much later, which is a huge problem,” Coetzee said.
And that delay in diagnosis can have devastatingly irreparable consequences. Children with Down syndrome have a higher risk of heart defects (some of which require urgent surgery), and need early therapy to overcome possible learning disabilities.
Down Syndrome South Africa’s chief Ancella Ramjas acknowledges that late diagnosis is indeed a widespread and pressing problem. In 2012, she visited Dr. George Mukhari Hospital in the country’s capital Pretoria.
“One of our objectives was to go into hospitals and look for new parents, so that we could support them in terms of information sharing,” she says.
But she did not expect to find more than 50 mothers waiting for her.
“Not one of them was there with a baby that was under eight months, and most of them had only found out about the diagnosis from eight months onwards. A lot of valuable time was lost,” she says. This can affect their chances of survival, and black babies are more likely to go undiagnosed compared to other race groups.
Artificial intelligence has also emerged as a viable option for helping those at risk of disease without the tools to know so. Christoffer Nellåker, a specialist who has pioneered techniques to detect diseases in people’s faces using machine learning, said a lot of this research has “tended to focus on Caucasian children from certain countries... But there is a huge need for this type of research to be done across populations around the world to make sure that the advantages for patients can be accessed in an equitable way.”
That inequity in health research has its roots in the colonial era, and one result of this is that there has historically been a greater investment in research on European populations than those in developing countries. Although there are efforts—such as Coetzee’s along with other large initiatives like Human Heredity and Health in Africa (H3Africa)—to promote research in and for African populations, there remains a knowledge gap between disease research, diagnosis, and treatment for different racial groups worldwide.
This is particularly true for South Africa, which is still trying to deal with the effects of its apartheid policy which segregated different races, and prevented black people from accessing services and socio-economic opportunities that were reserved to whites.
Practitioners in its beleaguered public health system often lack the training to identify conditions such as Down syndrome. This means that many black children born with conditions that require early diagnosis, treatment, and management, such as Down syndrome, are negatively affected.
Nellåker’s group at Oxford University has launched Minerva & Me, an online project encouraging people—particularly those with rare genetic diseases—to upload photos of themselves. This project, which calls for people to upload 2-D photos of themselves, is different from Coetzee’s database of 3-D images.
“There’s something like 7,000 different rare diseases,” he said, which can be either genetic and/or rare. Some of these diseases are so rare that they do not even have names, and experts in the field may only see them once in their career, if that. “We are trying to build models that will understand [markers] across all diseases” and be sensitive to diversity differences.
For Coetzee and her research team, the goal is more immediate: They want to help doctors and hospital staff diagnose genetic conditions as soon as possible. She returned to South Africa from the United Kingdom, and said that she thinks South Africa is where cutting-edge research on Down syndrome is being done. “I think I can make more of a difference here than overseas. So many people are doing what I’m doing overseas—I’m the only one doing this here.”