More accurate measurement of calories burned

Important points

• SEAS researchers have developed OpenMetabolics, an open-source smartphone-based activity monitor that uses machine learning and leg movements to estimate calories burned.
• Lab studies show that OpenMetabolics outperforms state-of-the-art smartwatches and fitness trackers.

While it may feel good to see your smartwatch tell you how many calories you burned after a workout, those numbers are often shockingly inaccurate, with an estimated error rate of 30% to 80%. The watch’s software makes its best guess based on variables such as heart rate, wrist movement, height, and weight, without actually measuring energy expenditure.

Biomechanics researchers at Harvard University have found a better way. New research from the lab of Patrick Slade, an assistant professor of bioengineering in the John A. Paulson School of Engineering and Applied Sciences (SEAS), introduces an open-source smartphone-based activity monitor called OpenMetabolics that uses machine learning to interpret muscle activity in a person’s legs into calories burned. In a laboratory study with human participants, the Harvard device was found to be twice as accurate as commercially available smartwatches and activity trackers. In addition to providing more accurate measurements of exercise, this study may also help scientists produce higher-quality research on the health effects of physical activity.

“Physical activity is very important for managing many aspects of health,” says Slade. “By relying on smartphone-based systems, this approach can be easily deployed for large-scale use and research studies, even in underserved areas.”

The research, published in Communications Engineering , was led by Ph.D. student Haedo Cho redeveloped a machine learning model that Slade’s group had previously shown could accurately extract energy expenditure values ​​from leg movements. The model uses continuous motion data captured by the smartphone’s gyroscope and accelerometer and interprets those swings and movements as values ​​for energy expended.

Previous iterations of lab activity monitors required heavily customized systems that were attached to two locations on a person’s foot. Cho aimed to redeploy OpenMetabolics across different types of people, movements, and activities solely through smartphone sensors. His research brings this technology closer to becoming a widely deployable commercial or high-quality research device.

Cho and his colleagues recruited 30 participants of various ages, sizes, and fitness levels to test their lab’s smartphone-based model against more common systems, such as those found in fitness trackers such as Fitbit heart rate models and pedometers. Participants wore the device and performed activities such as walking, cycling, and stair climbing.

Cho designed an experiment that captured real-life activity. “Most biomechanical studies that assess physical activity are done in the lab on treadmills…but this doesn’t capture how people walk in their daily lives,” Cho says. “People move at different speeds throughout the day. They might walk quickly to catch the bus. They might walk slowly to get groceries at Trader Joe’s. We emulated this type of scenario through voice prompts.”

Cho also created a “pocket motion artifact correction model” that maintains the accuracy of energy data even as smartphones bounce around in people’s pockets, in different styles of clothing, and at different angles.

Cho said he was particularly motivated to work on developing better ways to measure physical activity because while smartwatches and other measures are not common in many parts of the world, there are large disparities in population-level data in areas where smartphones are prevalent.

Additionally, as a marathon runner, he personally found the calorie information he saw to be wildly inaccurate. “I think we should do better at this, because there’s probably some mismatch between what people perceive and what the device tells them,” Cho said.

Slade added that the team, supported by a Harvard Impact Lab Fellowship, is actively exploring the use of technology to tackle global health challenges. Mr. Slade’s fellowship work focuses on understanding and addressing cardiovascular health risks in Latin American countries.

This research was also supported by the Harvard University Dean’s Competitive Fund for Promising Scholarships and the Raj Bhattacharyya and Samantha Heller Supportive Technology Initiative Fund.