advertisement
advertisement

PAID CONTENT

IN PARTNERSHIP WITH
IN PARTNERSHIP WITH

This AI-powered tech calculates a baby’s weight just from a video

Low birth weights are a serious, often undiagnosed problem. This app for rural health workers—a winner of Fast Company’s 2021 World Changing Ideas Awards—is more accurate than old, uncalibrated scales.

This AI-powered tech calculates a baby’s weight just from a video
[Photo: Wadhwani]
advertisement
advertisement

In 2015, 20.5 million newborn babies around the world were classified as having low birth weight. Babies weighing under roughly 5.5 pounds (2.5 kilograms) are 20 times more likely to die than their peers, and usually during the first month of their lives.

advertisement
advertisement

Low birth weight is an strong indicator of current and future health: Knowing a baby’s weight can give mothers and healthcare workers a chance for timely and simple interventions that could avert neonatal and infant mortalities, and help predict future health risks, including stunted growth, obesity, and diabetes. But the problem can be missed, especially in rural settings, simply because of inaccurate weight measurements, due to bad equipment and lax management.

In India, which has one of the highest rates of low birth weights in the world, about 25% to 30% of the cases are never identified due to poor gauging. Wadhwani AI, an “AI for social good” nonprofit based in India, is developing a new machine-learning weighing tool that’s designed to help the developing world easily complete these measurements without the need for reliable scales. It’s the winner of the Asia-Pacific category of Fast Company’s 2021 World Changing Ideas Awards.

[Photo: Wadhwani AI]

In rural India, the country relies on about 1.25 million government-recruited frontline healthcare workers, who “go home to home, tracking the progress of newborn babies and mothers,” says Rahul Panicker, Wadhwani AI’s chief research and innovation officer, until babies are about six weeks old. The visiting workers often have to use old-fashioned spring scales, which are low in supply and can be poorly calibrated. Weights are recorded by hand, and only weeks later inputted into central databases, at which point data is often lost or tampered with. The potential for human error is great.

advertisement
advertisement

Wadhwani AI’s mobile tool simply takes a video of a baby, and reconstructs a 3D model of the infant. Machine learning has trained the tool on parameters like body shapes and volumes, from which it can calculate the baby’s weight within seconds. The data—tamper-proof and geo-tagged with the baby’s location—is then uploaded directly onto medical record dashboards.

The tool can work as a feature on mHealth smartphone apps used by many healthcare workers in the developing world, so there’s no need for extra equipment or software. The workers can then recommend interventions—simple tasks mothers can carry out, such as careful temperature maintenance, and adapting breastfeeding and bathing regimens. The developers plan to soon configure the tool to gauge other important health indicators, like a baby’s head and upper-arm circumferences.

The company has been collecting data from hospitals and homes across four Indian states, including Gujarat and Telangana, via partnerships with state governments and nonprofits. In the early hospital trials, the tool has proven accurate. “We are able to detect low birth weight babies almost 90% of the time, with 90% accuracy,” Panicker says. The broader launch is planned for next year, with a national launch to follow, which will likely be followed by expansion to other developing countries.

advertisement

Wadhwani AI, which is also developing other AI tools in healthcare and technology, is concerned with modern technology not having purely good applications, especially in poorer countries. By leveraging AI in the developing world, Panicker hopes it can impact those “truly at the base of the pyramid,” he says. “We are hoping that we can bring the power of AI to help the truly underserved, so that it doesn’t stay restricted to optimizing Netflix recommendations.”