Though it might feel gratifying to complete a workout and see “calories burned” displayed on your smartwatch, that figure is often surprisingly inaccurate, with estimated error rates ranging from 30% to 80%. These devices rely on software that makes educated guesses based on factors such as heart rate, wrist motion, height, and weight, without actually measuring the energy expended.
Harvard biomechanics researchers have introduced a more precise alternative. A new study from the lab of Patrick Slade, assistant professor of bioengineering at the John A. Paulson School of Engineering and Applied Sciences (SEAS), unveils an open-source, smartphone-based activity monitor called OpenMetabolics. This tool employs machine learning to convert leg muscle activity into calories burned. A lab study involving human participants revealed that the Harvard device boasts double the accuracy of commercial smartwatches and activity trackers. This advancement not only promises more precise exercise measurements but also holds potential for enhancing scientific studies on health outcomes related to physical activity.
Revolutionizing Calorie Measurement
“Physical activity is critical for the management of many health aspects,” said Slade. “By relying on a smartphone-based system, this approach can be easily deployed for large-scale use and research studies, even in underserved areas.”
The study, published in Communications Engineering, was spearheaded by Ph.D. student Haedo Cho, who refined a machine learning model previously developed by Slade’s team. This model accurately extracts energy expenditure values from leg motion using continuous motion data from the smartphone’s gyroscope and accelerometer, translating these movements into energy expenditure values.
From Lab to Real-World Application
Previous versions of the lab’s activity monitor required a heavily customized system attached to a person’s leg in two places. Cho’s goal was to redeploy OpenMetabolics using only smartphone sensors, adaptable to various individuals, movements, and activities. His efforts bring the technology closer to becoming a widely deployable commercial or high-quality research device.
Cho and his colleagues recruited 30 participants of different ages, sizes, and fitness levels to validate the lab’s smartphone-based model against more conventional systems, such as those found in fitness trackers like the Fitbit heart-rate model or pedometer. Participants wore the devices while engaging in activities like walking, biking, and stair-climbing.
Real-Life Scenarios and Challenges
Cho designed experiments to capture real-life activities. “Many biomechanics studies evaluating physical activity are performed in the lab on a treadmill, but this does not capture how people walk in everyday life,” Cho explained. “People vary their speed during the day. When I catch a bus, I might walk fast. If I’m grocery shopping at Trader Joe’s, I might walk slowly. We emulated these types of scenarios through audio prompts.”
Additionally, Cho developed “a pocket motion artifact correction model,” which maintains the accuracy of the energy data despite the smartphone’s movement in pockets, across different clothing styles and angles.
“I think we should do a better job on this, because between what people perceive, and what the devices tell them, there is probably some mismatch,” said Cho, who is also a marathon runner.
Future Prospects and Global Impact
Slade highlighted that the team is actively exploring the use of their technology to address global health challenges, supported by a Harvard Impact Labs Fellowship. His fellowship work focuses on understanding and addressing cardiovascular health risks in Latin American countries.
The research received additional support from the Harvard Dean’s Competitive Fund for Promising Scholarship and the Raj Bhattacharyya and Samantha Heller Assistive Technology Initiative Fund.
As smartphones become increasingly ubiquitous worldwide, especially in regions where smartwatches are less common, this innovation could bridge significant gaps in population-level data on physical activity. The potential for large-scale deployment of OpenMetabolics could revolutionize how individuals and researchers alike understand and manage physical health.
