A pocket-sized gait lab
Shrek, Gollum, or the Hulk: “motion capture” is a widely known technology in movie-making. The bodies of live actors are equipped with small markers; cameras record their movements, which are later fed into an animated character. The same technology that brings fictional creatures to life in Hollywood helps physicians to precisely measure human movement.
In Brian Horsak’s gait laboratory, 16 cameras are mounted on the walls. They capture with great precision how a person walks: how the foot is placed on the ground, how much the knee bends, how the hip moves, and how the pelvis tilts. “You can picture it as a three-dimensional X-ray without radiation,” says Horsak, a researcher at the St. Pölten University of Applied Sciences who has been working on motion analysis for years. In his FWF-funded project “Out-of-the-lab Biomechanical Assessments in Cerebral Palsy,” Horsak and his team are investigating whether part of this analysis could in the future be conducted outside of specialized laboratories with the help of smartphone videos and artificial intelligence (AI). The project focuses on children and adolescents with cerebral palsy.
About the project
Cerebral palsy, a motor disability that develops in childhood, impairs the ability to walk. Gait analyses are important supportive measures for therapy, but they are costly and time-consuming. Researchers are now developing an analysis method using smartphones and AI to provide additional movement data and improve treatments.
When walking has a story to tell
Cerebral palsy, a neurological movement disorder, is considered the most common motor disability in childhood affecting approximately two to three out of every 1,000 newborns. Resulting from brain damage that occurs in an early phase of life, cerebral palsy can have a wide variety of consequences. Some children can walk independently, while others need supporting devices or a wheelchair. Muscle tone, coordination, balance, and movement control are frequently impaired.
For the children affected and their families, the diagnosis often brings many questions and decisions: what therapy helps? Are assistive devices needed? How will walking ability develop as the child grows? Will the child need surgery?
Growth plays a particularly important role. Children get taller and heavier, and their muscles and bones undergo changes. Some impairment that a child was able to make up for adequately for a while may become more challenging later as a teenager. In the worst scenario, the walking ability of patients deteriorates significantly.
This explains the importance of understanding the gait of those affected in detail. Observing with the naked eye, one might notice that a child walks unsteadily, stiffly, or asymmetrically. But the human eye misses many details. “Our eyes can only capture 25 frames per second. That’s not enough in this case,” notes Brian Horsak. Movement is something that occurs quickly and involves more than just one joint. The knee, hip, pelvis, and ankle are all interconnected. If one area changes, it often affects the others. “That’s why we need to record and quantify movement,” emphasizes Horsak.
What the lab sees
In a traditional gait lab, patients walk a short distance while multiple cameras track markers attached to specific points on the body that indicate the spatial positions of the knees, hips, or shoulders. The system detects their positions to the nearest millimeter. Force plates embedded in the floor also measure the forces acting at each step. Combining the camera data with the force measurements enables the researchers to calculate the torques in the joints.
A few steps thus generate curves, angles, and models. In this way, doctors can see aspects such as how the right and left legs differ, how much a knee remains bent, or whether the pelvis tilts forward. These data help in planning therapies, preparing for surgery, or evaluating treatment outcomes.
Children with cerebral palsy often exhibit what is known as a “crouch gait”, a walking pattern involving excessive bending at the knees and hips. From the outside, this may simply look like a strained gait, but the lab analysis reveals a detailed trajectory of joint angles and movements. This is precisely where the value of such laboratories lies: they make visible what might otherwise be overlooked.
A detailed look
As helpful as these data are, collecting them is still a complex process. A gait lab requires at least ten cameras, specialized software, force plates, and trained staff. The markers must be precisely attached to specific anatomical points, and the data must be analyzed and interpreted by physicians. “A lab like this can easily cost 500,000 euros or more,” notes Horsak. For this reason, gait labs are primarily found at specialized orthopedic clinics, rehabilitation centers, or research institutions. In his project, Horsak is collaborating with several institutions, one of them being the Orthopedic Hospital Speising in Vienna, where children with cerebral palsy are regularly examined.
Given that the number of gait laboratories is limited, access remains difficult for many families. Travel distances can be long, appointments time-consuming, and examinations expensive. In practice, gait analysis is often performed only when major therapeutic decisions or surgery are imminent.
But what happens in the meantime is often less well known: how does mobility change in everyday life? Will any deterioration be noticed early enough? Is mobility declining? Should the medical team take a closer look? This is where Horsak’s project comes in.
Movement captured on video
Here is the idea: a child takes a few steps, a smartphone records the movement, and AI analyzes the video. The images are used to create a three-dimensional motion model – similar to what’s done in a gait lab, but without markers, without a camera system, and without expensive infrastructure. “We’re trying to put the entire gait lab in the pockets of physiotherapists, doctors, and families,” says Horsak.
Horsak emphasizes that the goal is not to replace the traditional gait lab. Complex surgical planning still requires high precision and specialized labs. The smartphone analysis is intended to help in situations where no motion data has been collected in the past – between two lab appointments, for instance, at home, during physiotherapy, or in countries where access to gait labs is limited.
When AI needs training
In order for this to work, however, the AI involved must be tested more thoroughly and upgraded. While many AI models are already trained using large datasets, these do not automatically include the body shapes and movement patterns that can be typical of cerebral palsy: bent knees, legs turned inward, altered posture, bone deformities, and asymmetrical movements. “The AI already knows a lot, but not everything,” says Horsak. That’s why it still “hallucinates” from time to time: when encountering movement sequences that the AI doesn’t recognize, it delivers inaccurate or, in some cases, fictional body positions.
Hence, Horsak’s project isn’t simply about applying existing AI to medicine. The first step is to test its limits. The team compares the results of the smartphone analysis with the gait lab gold standard. This reveals which joints, directions of movement, and specific questions the method already handles well – and where it has deficiencies.
In a next step, the models will be improved. To this end, the team is collaborating with the Speising Orthopedic Hospital to collect video data and motion analysis data from children with cerebral palsy. These data can be used to retrain and fine-tune existing AI models: nutshell: a model that already has a basic understanding of human movement is designed to learn how to recognize better the movements of children with cerebral palsy.
What has already been achieved
As a first step, Horsak’s team worked with healthy participants. They were asked to imitate various gait patterns that can also occur in children with cerebral palsy. These movements were recorded using smartphone video and a traditional gait lab at the same time.
Horsak found the results surprisingly promising. The method worked particularly well for large movements such as flexing and extending of the knee or hip. In those cases, the deviations averaged about five degrees. This is close enough to be useful for certain clinical questions.
But he also found that not everything works equally well. Some joints and directions of movement are more difficult to capture. Small movements have a greater impact. And there are some individuals for whom the model works significantly better than for others.
Horsak and his team want to find out why. Currently, they are working directly with children with cerebral palsy. Nearly 30 participants have already been enrolled, with more to follow. According to Horsak, the initial results look just as promising. On the other hand, there is still one central question: how reliable is the method really – and for which applications is the accuracy it provides sufficient?
Horsak’s grand vision is for “pocket-sized motion analysis” to be ready soon for use by doctors and physiotherapists as well as by the people sharing the patients’ lives. The ideal outcome would provide additional information, improve therapies, and make life easier for patients and their families. “I see so much potential there to do good for the families concerned.”
About the researcher
Brian Horsak became interested in motion analysis through sports studies. Today, he conducts research at the University of Applied Sciences St. Pölten (USTP) at the intersection of biomechanics, health technology, and AI. In his project “Laboratory-Independent 3D Gait Analysis in Cerebral Palsy,” funded by the Austrian Science Fund (FWF), he is investigating how reliably markerless 3D gait analyses based on smartphone videos work in children with cerebral palsy. The project runs from 2024 to 2027.
Publication
Validity and reliability of monocular 3D markerless gait analysis in simulated pathological gait: A comparative study with OpenCap, in: Journal of Biomechanics 2025
