IMU-Based Hip Adduction and Flexion Moment Estimation for Osteoarthritis and Healthy Individuals During Normal and Modified Gait

For millions of people living with hip osteoarthritis, the simple act of walking is a constant negotiation with pain. The condition wears down the smooth cartilage that cushions the joint, turning every step into a grinding friction that can lead to severe disability. While doctors can see the damage on an X-ray, the true story of the disease unfolds in the invisible forces acting on the body while a person moves. Every time a foot hits the ground, the hip joint absorbs a massive load, and the way a person walks—how they lean, how they turn their toes, how fast they move—determines exactly how much stress that joint endures. Over time, these cumulative forces can accelerate the disease, yet for most patients, this dangerous loading happens in the quiet of their daily lives, far away from the specialized clinics where such measurements are usually taken.

Until now, understanding these forces has required a trip to a biomechanics laboratory. There, patients walk across a floor embedded with giant pressure sensors while cameras track their movements with laser precision. This setup provides a detailed snapshot of joint stress, but it is expensive, cumbersome, and captures only a few seconds of walking in a controlled environment. It cannot tell a doctor how a patient's hip behaves during a morning walk in the park, a grocery run, or a moment of distraction. The challenge for medical science has been to find a way to bring this level of insight out of the lab and into the real world, using small, wearable devices that can track the body's motion without the need for a room full of equipment.

A team of researchers from Shanghai Jiao Tong University, Kyoto University, and the Chinese University of Hong Kong has taken a significant step toward solving this problem. They developed a new method to estimate the twisting and bending forces inside the hip joint using only small sensors attached to the body. These sensors, known as inertial measurement units, are tiny devices that can detect acceleration and rotation, similar to the chips found in smartphones. The researchers wanted to know if they could use these simple signals to reconstruct the complex forces acting on the hip, even for people with arthritis and even when those people changed the way they walked.

To test their idea, the team recruited eighty-four participants, including young healthy adults, older healthy adults, and people diagnosed with hip osteoarthritis. They asked everyone to walk on a treadmill and across a floor equipped with force sensors while wearing eight of these small devices on their trunk, pelvis, and legs. The participants did not just walk normally; they were asked to modify their gait in specific ways to see if the system could keep up. They walked faster and slower, turned their toes inward and outward, widened and narrowed their steps, leaned their torsos to the side, and even tried to walk while counting backward by threes to simulate a distracted mind. This created a vast library of movement data, capturing how the hip behaves under a wide variety of conditions.

The researchers then built a computer model, a type of artificial intelligence designed to learn patterns from data. They trained this model to look at the signals coming from the sensors on the trunk, pelvis, and legs and predict the corresponding forces inside the hip. The goal was to create a system that could work for anyone, not just the people it was trained on. Initially, they tested a "one-size-fits-all" version of the model. This version was reasonably good, but it struggled to get the exact numbers right for every individual, especially for those with arthritis whose movement patterns differed significantly from healthy people.

The breakthrough came when the researchers introduced a technique called "few-shot personalization." Instead of trying to force every person to fit the same mold, they allowed the model to learn a little bit about each new person using just a tiny amount of data. Before testing a new participant on the difficult, modified walking tasks, the system asked them to walk normally for just a few steps. The model used these few steps to adjust its internal settings to match that specific person's body size, how they wore the sensors, and their unique way of moving. It was a process of calibration, similar to tuning a radio to a specific station, but done in seconds by the computer.

The results showed that this small adjustment made a huge difference. When the model used only the data from a few normal steps to personalize itself, its predictions became remarkably accurate. For the forces that push the hip inward and the forces that bend it forward, the error dropped significantly, and the predicted movement patterns matched the real measurements with a high degree of precision. This accuracy held true even when the participants switched to the difficult tasks, such as walking with a wide step or leaning their trunk. The system did not need to be retrained for every new walking style; the initial calibration was enough to let it generalize to new situations.

Perhaps most importantly, the system worked just as well on an independent group of people who had never been part of the training process. This group included more people with hip arthritis and developmental issues with their hips, walking on a flat surface rather than a treadmill. The personalized model successfully estimated their hip forces, proving that the method could transfer across different environments and different types of patients. The researchers found that the system could accurately capture not just the average force, but the specific peaks of stress that occur at the moment the foot hits the ground, as well as the total amount of stress accumulated over the entire step. These details are critical for doctors, as they indicate which walking patterns might be harmful and which might help protect the joint.

The study also explored how many sensors were actually needed. They found that using sensors on the trunk, pelvis, thigh, and shin provided the best balance between accuracy and comfort. Adding a sensor to the foot did not improve the results enough to justify the extra burden on the patient. This suggests that a practical, wearable system for daily use could be quite simple, requiring only four small devices rather than a complex array of equipment.

The implications of this work extend beyond the laboratory. By enabling accurate, continuous monitoring of hip loading outside of a clinic, this technology could help doctors identify harmful walking habits before they cause further damage. It could allow patients to receive real-time feedback on how to adjust their gait to reduce pain and slow the progression of their disease. Instead of waiting for a yearly checkup where a doctor might only see a static image of the joint, patients could wear these sensors during their daily lives to see how their bodies respond to different activities. The researchers suggest that this approach could eventually lead to personalized rehabilitation plans that are adjusted based on real-world data, rather than theoretical models.

While the study was conducted under controlled conditions with participants walking on treadmills and marked floors, the success of the personalized model in an independent group walking over ground suggests a strong potential for real-world application. The researchers acknowledge that future work will need to test the system in completely uncontrolled environments, such as busy streets or uneven terrain, and explore ways to reduce the need for the initial calibration steps. However, the core finding remains clear: by combining wearable sensors with a smart, adaptable computer model, it is possible to see the invisible forces acting on the human hip, opening a new window into the management of one of the most common and debilitating joint diseases of our time.