Table of Contents
A Powerful AI Tool Could Help Medical Professionals Treat Serious Motor Dysfunction
Motor dysfunctions accompany a variety of health conditions, including neurological disorders like Parkinson’s disease and cerebral palsy, and events like stroke and brain trauma. These kinds of dysfunctions often affect walking and balance, and treating them depends on an accurate picture of the way in which a patient’s gait, or pattern of walking, differs from that of a typical “healthy” person. Gait analysis is an essential tool for diagnosing and treating serious motor dysfunctions as well as other types of movement problems. But obtaining a detailed and accurate gait analysis typically requires the services of specialists, and the process can be time-consuming and expensive.
Diagnosing Motor Dysfunctions: A New Approach
Currently, the best strategy for diagnosing motor dysfunctions through gait analysis is a technology called “motion capture,” which is widely used in film and television to set up CGI scenes. But the motion capture technologies currently available for diagnosing motor dysfunctions typically require tools and expertise that can put them out of reach for many neurologists, orthopedists and other professionals working with people who have serious motor dysfunctions.
Now, though, a recent study done jointly by Stanford University and Gillette Children’s Specialty Hospital in Minnesota reveals a new motion-capture approach to video-assisted gait analysis, or VOGA, that promises to put sophisticated and accurate gait analysis technology in the hands of any health professional.
Working with patients who have cerebral palsy, the Stanford/Gillette team combined single camera motion capture with the data processing power of an advanced artificial intelligence platform called OpenPose. These tools allowed researchers to develop a gait analysis tool capable of producing results in a quarter of the time required by standard motion capture technologies – and it can work with video taken at home and uploaded for analysis.
What Causes Motor Dysfunction?
Motor dysfunctions (also called movement dysfunction) are part of a broader group of impairments called movement disorders –a term that covers a variety of issues that affect the normal movement of the limbs or body. Motor dysfunctions can be caused by health conditions that affect the nerves and nerve signals sent by the brain and spinal cord, and by problems with the muscles, bones and joints. Movement disorders can include difficulties with all kinds of motions, including bending, sitting, and raising the arms or turning the head. But most motor dysfunctions involve impairments with walking and balance.
A number of neurological diseases can cause varying degrees of motor dysfunction. Parkinson’s Disease and multiple sclerosis affect the nerves that direct motion by sending signals to the spinal cord and limbs. Alzheimer’s disease and other forms of dementia that affect neural connectivity in the brain can also cause changes in the way a person moves.
Damage to various parts of the brain can also cause motor dysfunction. Stroke, brain tumors, or brain injuries and infections can damage parts of the brain that control movement and balance. Cerebral palsy, a complex set of disorders caused by damage to an infant’s developing brain, accounts for many serious motor dysfunctions involving standing, walking and balance.
Treatments for motor dysfunctions depend on the cause. Some, such as motor deficits caused by diseases like Parkinson’s, can be managed with medications. In other cases, physical therapy and other rehabilitation programs can help to improve motor function. But treating motor impairments depends on an accurate diagnosis – and gait analysis is an important tool for that.
Diagnosing Dysfunction Through Patterns of Movement
Diagnosing a motor dysfunction and planning treatment begins with getting a clear picture of the way an impaired person’s gait differs from the “norm” of healthy individuals. With video assisted gait analysis, neurologists and other specialists can use video capture to evaluate a person’s movement, typically measured in three dimensions: walking speed, walking cadence (the number of steps per minute) and symmetry – a relatively equal distribution of movement on both sides.
Capturing those three dimensions on video and completing an analysis typically requires a complex camera set-up and access to a large storehouse of video data that is used as a reference for evaluating deviations in movement. Until relatively recently, that kind of processing had to be performed manually by a team of trained professionals — a process that could only be done at specialized locations and could take several hours.
Those constraints make it difficult for smaller clinics and independent professionals to use VOGA directly in their practices. Using a single camera, such as the kind most people use at home, could save both time and money, but those cameras can capture movement only in two dimensions. And software for quickly analyzing patient video and a large body of comparison data would save time and eliminate the need for specialists. The study conducted by Stanford and Gillette aimed to create just such a solution, making it possible for professionals everywhere to diagnose serious motion dysfunctions quickly and easily in their own practices.
With the help of an AI-powered platform called OpenPose, which is capable of processing a database of more than 1.5 million images of people in motion, the Stanford/Gillette team was able to detect key points of impairment in 1700 videos of cerebral palsy patients using just a single camera. Their work represents a major achievement that, according to the journal Nature, will “democratize the study of neurological and musculoskeletal disorders” by making it possible to conduct a detailed evaluation of motor dysfunctions more quickly and economically than ever before.
Artificial Intelligence Democratizes Diagnostics
Artificial intelligence is a branch of computer science dedicated to creating “smart” machines capable of performing tasks that humans usually do. In a process called machine learning, the algorithms that power intelligent machines learn to make conclusions, predictions and decisions independent of human input.
Artificial intelligence is becoming an essential tool in medicine, largely for its ability to find patterns, identify anomalies and extract insights from increasingly large databases of medical information stored in the cloud. Because AI applications are capable of finding complex patterns humans may not be able to detect, these tools can also discover problems that might otherwise be missed.
Once trained, AI-powered tools can perform operations on the data sets of medicine’s “big data” banks faster and more accurately than human operators can. Using the deep neural networks of advanced machine learning, these tools can learn what a “normal” gait looks like and then measure deviations captured on video against a massive database of videos showing a wide range of variations in human gait patterns.
These capabilities provided the framework that helped the Stanford/Gillette researchers make their gait analysis model work. With the help of OpenPose and similar technologies, researchers could upload video images and compare them to thousands of stored files to identify impairments in the speed and cadence of a patient’s gait.
The team’s use of a single camera for motion capture was also an essential step toward “democratizing” VOGA technologies for general use. Without the need for a dedicated 3D camera arrangement, the single camera approach can even make it possible for patients or their families to record video at home and upload it to a website for analysis, which eliminates the need for an in-person clinic visit.
This system could be implemented in any clinic using AI-powered software installed on office computers, and requires no special training beyond learning how to upload video and run the program. Results can be returned in minutes, not hours, with detailed information on a variety of customizable parameters such as knee flexion and walking speed. Any healthcare practitioner can interpret the results for quick diagnosis and treatment planning.
The Future of AI for Treating Motion Dysfunction
The Stanford research opens doors to faster, more accurate diagnosis of motion dysfunctions and puts new tools at the disposal of healthcare professionals of all kinds. But the Stanford team cautions that their work is not without limitations.
At this point, a 2D camera that records video only from the side still lacks the ability to capture a fully three-dimensional picture of a person’s gait. And patients need to be able to match database videos in terms of clothing and the type of movements they do, since the AI platforms used in the study are not yet able to read data that deviates too much from their learned patterns. The study’s runners point out that the next generation of OpenPose and other gait analysis technologies may be able to extract 3D data from single camera captures, and perform operations on even bigger databases – all without a significant increase in costs or a decrease in efficiency.
For now, though, the Stanford/Gillette study and similar research reveals another way in which the tools of artificial intelligence, machine learning and the cloud are transforming the way healthcare is developed and delivered. With quicker, faster and more accurate tools for diagnosing and treating serious motion disorders, professionals across the healthcare spectrum have more options for helping patients and their families manage a wide range of conditions that affect movement and balance.