Researchers at Georgia State University are making breakthrough strides in stroke rehabilitation for millions of Americans by leveraging an unexpected tool: math.

By applying advanced data analytics and machine learning, the team is transforming monotonous repetitive movement therapy into a fun, interactive and engaging activity via a first-of-its-kind stroke recovery robot for the hands and feet. The technology – a product of healthcare robotics leader Motus Nova – encourages greater therapeutic compliance and is resulting in better patient outcomes. It is also now widely accessible with newly-announced health insurance coverage.

“Literature shows that the more repetitive motions stroke survivors do, the likelier they are to regain use of their limbs,” said Russell Jeter, an assistant professor of mathematics at Georgia State University who previously served as Director of Analytics and Software Engineering at Motus Nova. He is now working to solve complex mathematical challenges to further refine the technology as it ramps up for large-scale use. 

Jeter’s work was recently featured in SIAM News, a publication of the Society for Industrial and Applied Mathematics (SIAM), and presented at the 2024 SIAM Conference on Mathematics of Data Science, which took place in Atlanta in October. 

Nearly two-thirds of the estimated seven million stroke survivors in the U.S. require rehabilitation services after hospitalization to improve mobility and dexterity in their limbs. But until now, the only option for stroke therapy was often an expensive, outpatient program of weekly sessions that rely on patients’ continued completion of mundane movement exercises at home.

The patented assistive technology consists of exoskeletons that are strapped to the wrist or foot and supported by a pneumatic air pump that acts like a muscle, offering assistance when a user is struggling to carry out a movement or resistance when they need to be challenged. As users play computer games (modeled after popular video games like Guitar Hero, Pong, Solitaire and Space Invaders), they control what happens on the screen by making small hand and foot motions. Sensors on the devices detect when those movements are correct, providing biofeedback to the brain to help rebuild critical neuro pathways.

For example, when a user is wearing the exoskeleton, flexing a wrist or foot upwards or downwards might correspond to moving a spaceship right or left. “When you’re willing your body part to move in a specific way and you mentally associate that movement with something you see on a computer screen, even if you aren’t able to do the movement, your brain will encourage new neuro pathways to grow,” Jeter said.

Jeter’s work shows how Georgia State mathematicians improved the robots’ collective ability to autonomously classify a user’s range of motion, an important breakthrough that supports wider adoption of the technology. Being able to estimate range of motion in three broad categories — no movement, low movement or high movement — allows clinicians to better align the robotic therapy with patient needs.

“Think of exercises as stretching, gross motor control or fine motor control,” Jeter said. “If someone doesn’t have a lot of movement in their hand or foot, it would be inappropriate to assign them a fine motor exercise. But if they are progressing to a higher level of function, you would want to increase the level of difficulty.”

The highly sensitive robots collect roughly 30 data points per second. Using the summary data from 33 patients over the course of 30-minute therapy sessions in a clinical setting, Jeter’s research team identified the best mathematical method for detecting residual stroke severity in terms of no, low or high range of motion. The resulting model — a decision tree method called light gradient boosting — achieved 96% accuracy and was especially successful at classifying patients with a low range of motion.

This outcome is important as the technology scales because it means that a doctor is no longer required to classify residual stroke severity. Instead, lower-level technicians can do so in a clinical setting so that more patients can effectively be treated in less time. 

“Simply stretching an exercise band over and over again is tedious and gets tiring,” Jeter said. “But when you can achieve the same result through playing an interactive computer game that tracks your progress, it’s extremely motivating.” He noted that 82% of people who use the innovative robots report improvement in motor control.

Originally piloted as an innovative at-home solution during COVID-19 lockdowns, the technology has now been used by more than 4,000 patients and is preparing for a larger scale launch — including in clinical settings.

Moving forward, math continues to play an important role as the team adjusts the equipment based on lessons learned from user wear and tear. For instance, they are developing better reporting systems so users can visualize their progress over time and working to keep games interesting so patients remain engaged.

Ultimately, the accurate representation of user movements is a significant mathematical challenge that requires balancing the trade-off between sensor accuracy, speed of processing and bottom-line cost. “Being able to see real impact on real people is incredibly exciting,” said Jeter. “There are a lot of issues that need to be solved to keep this technology affordable and you need math to do that.”