Portable movement test uses artificial intelligence to detect early signs of cognitive decline

A new study published in Alzheimer Disease & Associated Disorders provides initial evidence that a portable and affordable device can accurately identify older adults with mild cognitive impairment based on how they move during everyday tasks. Using a combination of a depth camera, a force plate, and artificial intelligence, the system was able to correctly classify 83% of participants with mild cognitive impairment. The results suggest this tool could be used to expand access to early screening, especially in communities with limited resources.

Mild cognitive impairment refers to changes in memory and thinking that are noticeable but not severe enough to interfere with daily life. It often represents a transitional stage between normal aging and more serious conditions such as Alzheimer’s disease or other forms of dementia. Early detection is important because treatments that may slow progression—like the new drug Lecanemab—are only approved for people in the early stages of the disease.

However, getting an accurate diagnosis is often a long and expensive process that requires access to specialized professionals. In rural or underserved areas, these evaluations are especially hard to access. Only a small percentage of older adults with mild cognitive impairment receive a formal diagnosis, making early intervention difficult.

Researchers at the University of Missouri wanted to find a way to bring screening tools directly into community clinics and homes. They developed the Mizzou Point-of-care Assessment System, or MPASS, which is a lightweight and portable device that includes a depth-sensing camera and a custom-built force plate. This setup allows for detailed measurements of how a person moves while walking, standing, and performing other functional tasks.

“Our original goal was to develop accessible technologies for assessing movement and balance for use in the clinic. We have a very nice traditional gait lab with gold standard equipment for measuring human movement (motion capture, force plates, EMG). However, this system was rarely used outside of research projects. These systems are just too expensive and too complicated for everyday use in clinics or other facilities outside the lab,” explained study author Trent M. Guess, the director of the Mizzou Motion Analysis Center.

Free daily newsletter

“In 2020, we received funding from the University of Missouri to develop MPASS. Our initial target for the MPASS was concussion assessment and found that the platform could distinguish persons in the acute concussion phase as well as identify lingering effects of concussion on movement and balance. With these promising early results, we wanted to know if the MPASS could detect the effects of mild cognitive impairment on movement and balance.”

“The connection between gait, especially during dual tasking, and cognitive decline is well known. Alzheimer’s is a truly devastating disease, like many others, I have family members and close friends who have had their lives turned upside down by Alzheimer’s. It is rewarding to be able to work on a technology that may be able to help detect dementia in its earliest stages.”

For the study, the team recruited 47 participants, all over 60 years old. Nineteen had been diagnosed with mild cognitive impairment, either through a prior evaluation at a neuropsychology clinic or based on their score on the Montreal Cognitive Assessment, a standardized cognitive screening test. The other 28 participants had no known cognitive issues and served as the healthy comparison group.

Each participant completed a series of motor tasks while being observed by the MPASS system. These tasks included standing still, walking a short distance, and standing up from a seated position. To make the tests more challenging—and to better reveal signs of cognitive decline—participants had to do each task while counting backward by sevens from a random number between 70 and 100. This type of “dual-task” test places extra demand on both attention and coordination, making it more likely to reveal subtle cognitive deficits.

The MPASS device recorded data using both the depth camera and the force plate. The camera tracked body position and joint movements in three dimensions, while the force plate measured how the person shifted their weight and maintained balance. The researchers extracted 27 different variables from these recordings, including stride length, time to complete tasks, and how much a person swayed while standing still. Some of the data was captured with participants’ eyes open and some with eyes closed, to test the role of visual input in balance.

All of this data was then analyzed using three types of machine learning models: logistic regression, support vector machines, and decision trees. These models are designed to recognize patterns in large data sets and make predictions based on those patterns. The models were trained on most of the participant data and then tested on a smaller group to assess how well they could identify which individuals had mild cognitive impairment.

The decision tree model turned out to be the most accurate, correctly identifying 83% of participants with mild cognitive impairment. It also achieved a perfect score for specificity, meaning it correctly recognized all healthy individuals as not having cognitive impairment. The machine learning model found that the most important clues came from balance-related measures, particularly when the person was asked to stand still with their eyes closed while doing math out loud. Five out of the top six predictive features came from measurements of balance, such as how much a person’s center of mass swayed while standing. The remaining key feature was stride length while walking.

Interestingly, measurements from the sit-to-stand task did not contribute much to the final model, even though this test is often used in clinical settings to assess strength and mobility. The researchers suggest that future studies might still explore more advanced ways to analyze this task, since their version included motion data not normally captured in traditional assessments.

“We were thrilled to learn that the MPASS could detect subtle signatures in movement associated with mild cognitive impairment. Currently, mild cognitive impairment is grossly underdiagnosed. One study estimated that only 8% of older Americans expected to have mild cognitive impairment receive a clinical diagnosis. An efficient, inexpensive, and accessible method for mild cognitive impairment screening would be very beneficial in the fight against Alzheimer’s and other dementias.”

“The MPASS measures multiple aspects of motor function (e.g. static balance and gait) and combines cognitive and motor tasks (e.g. walking while solving math problems) to provide more sensitive data for detecting motor function changes associated with cognitive decline. MPASS assessments generate diverse data sets and the use of artificial intelligence can detect intricate relationships in this data, providing a means for instantaneous diagnosis.”

However, the authors acknowledge some limitations. The sample size was small, with only 19 participants in the mild cognitive impairment group. The participants were also not very diverse in terms of race or geographic background, so future studies will need to include a wider range of individuals to ensure the findings are broadly applicable. Some data was also lost during testing due to issues with body tracking, although the research team has since refined their procedures to avoid this problem in future work.

Despite these limitations, the results suggest that a portable, low-cost system like MPASS could be a practical tool for early detection of cognitive problems, especially in settings where access to specialized testing is limited. Because the device is easy to use and doesn’t require blood tests or imaging, it could potentially be used in primary care offices, senior centers, or even in people’s homes. This could help identify people at risk earlier and connect them with interventions while treatments are most effective.

The research team is now working on expanding the study with funding from the National Institutes of Health. They plan to include more complex walking tasks and evaluate other types of movements to further improve the system’s accuracy. The researchers believe that combining cognitive and motor testing with artificial intelligence holds great promise for improving screening and outcomes for older adults.

The study, “Feasibility of Using a Novel, Multimodal Motor Function Assessment Platform With Machine Learning to Identify Individuals With Mild Cognitive Impairment,” was authored by Jamie B. Hall, Sonia Akter, Praveen Rao, Andrew Kiselica, Rylea Ranum, Jacob M. Thomas, and Trent M. Guess.