Can AI Tell if a Child Is Malnourished?
A new program may be able to spot malnutrition in a simple photo, making it easier to assess nutrition problems in volatile regions
In war, disaster or famine-stricken areas, charities often report the number of children suffering malnutrition in terms of percentages—one in 10, one in five, one in three. But did you ever wonder how these rates are calculated?
Discerning who and how many people are malnourished takes a fair amount of professional skill. Unfortunately, areas experiencing humanitarian disasters often lack workers with this kind of training.
Now, a Kenya-based company has developed an AI that may be able to judge a child’s nutrition status through a simple photo. They hope the technology, called MERON (Methods for Extremely Rapid Observation of Nutritional status), can help collect vital data in areas where trained workers are unavailable or impractical.
“We work in highly insecure areas where we provide remote monitoring services,” says Ben Watkins, CEO of Kimetrica, a company whose mission is to help governments and non-profits increase the impact effectiveness of aid money. “So we’re monitoring the food security and nutritional situation in areas that agencies don’t necessarily have very reliable access to.”
MERON has been trained on a database of photos to be able to recognize the facial features, like the roundness of cheeks, that correspond to malnutrition. All it needs is a facial photo and it can instantly categorize the image as normal, moderately malnourished or severely malnourished. Initial trials suggest the AI has a 78 percent accuracy rate in detecting normal weight individuals; Kimetrica is currently working on trials with photos of malnourished children.
The idea for MERON came from Watkins’ teenage daughter. Watkins and his team had been discussing ideas for simple, less-invasive ways of assessing child malnutrition. His daughter said, “Why don’t you just take pictures of people’s faces? You can tell how heavy people are by just looking at their faces."
It was a good idea, Watkins thought. After all, facial features are one factor trained human assessors use in visually judging malnutrition. His daughter’s name? Meron. The AI’s name is actually a backronym for its originator.
There are currently several methods of assessing acute malnutrition in children. A trained observer can make a visual assessment based on factors like muscle wasting. Assessors can measure a child’s mid-upper arm circumference – the cutoff for “severe acute malnutrition” is 11 centimeters for children under five years of age. Or a weight-height ratio can be used.
Measuring the extent of malnutrition is critical both for getting aid money and for deciding which children are in need of emergency medical treatment and therapeutic foods – often energy-dense pastes fortified with micronutrients.
But Kimetrica often works with highly volatile, highly remote areas. Even if trained assessors are available, the work is often dangerous to both them and the families they’re assessing. Local authorities in war-torn regions may not appreciate that international agencies are raising awareness of their internal chaos. Setting up a tent to take arm or height and weight measurements may draw unwanted attention.
“There’s a need for discreet technology where it can be used without raising awareness or being too conspicuous in the field,” Watkins says. “The idea of using a smartphone is appealing in that respect, because you can quickly take a snap.”
Andrew Jones, a public health nutritionist at the University of Michigan, agrees that current methods of assessing malnutrition can be invasive in certain contexts. Measuring arm circumference may involve removing clothing, which can be taboo in some cultures. And getting height measurements requires training and a child’s cooperation.
“It’s actually quite traumatic for some little kids to have a stranger come and take their height,” Jones says.
Jones says he can see the role for technologies like MERON in humanitarian emergencies.
“In those contexts I can certainly see potentially a need for screening lots of kids in a short period of time with limitations on trained staff,” he says.
Jones notes that severe acute malnutrition – the kind that presents itself with wasted limbs and swollen bellies – is actually much less common than other forms of malnutrition. More common is “stunting” – the impaired growth and development that can come from poor diets. Stunted children are not necessarily skinny – some in fact look quite plump – but they may suffer cognitive impairments and poor health.
“There are many more stunted children in the world than there are children who are severely acutely malnourished,” Jones says. According to data from the WHO and UNICEF, about 155 million children worldwide are stunted, while some 16 million suffer severe acute malnutrition.
Kimetrica has been field-testing MERON, and has a few kinks to work out before the program might be unrolled. First, the photos used must feature the child facing forward, in good light. This requires some training on the part of the photographer, whether a parent or a local worker. Second, MERON must be tested on children of different nationalities and ethnicities, to make sure it’s equally accurate for everyone. The team will then need to create a seamless app that gives instant feedback.
Watkins hopes MERON might eventually have applications beyond severe acute malnutrition, such as diagnosing diseases like kwashiorkor, a form of protein malnutrition that causes swelling, or even assessing obesity rates.