Neurologist Heidi Schambra from New York University describes the post-stroke brain as “avid to learn”: in the initial weeks of recovery, the brain is ripe to reform its connection to muscle and regain abilities that once came naturally. “There’s a window of opportunity where we can really engage the brain,” Schambra said. “There’s this heightened plasticity that the stroke itself sets into action.”
For many recovering patients, rehabilitation involves working on some form of impaired arm mobility—on the series of motions the limb strings together to brush one’s teeth or butter one’s toast for breakfast. As primed as the brain may be to get going on these motions, rehabilitation researchers and therapists are still unsure as to how many of them patients should try making during rehab sessions. That’s because we don’t yet have a practical and precise method for counting and identifying motions made in a given session, Schambra said.
Without clarity on how much training a patient is getting, it’s hard to figure out how to tweak it. In a new study published this week in PLOS Digital Health, Schambra and other colleagues at NYU outline a new approach they’ve developed to pin down that data, using AI to count and identify arm motions about 370 times faster than human observers can. It may be able to help stroke patients get closer to the right quantities of motions needed to jumpstart their recovery.
To pull everything together and pioneer this approach, the researchers filmed 41 recovering stroke patients with high-definition video cameras as the patients did typical rehabilitation activities for arm mobility—like pouring a glass of water, moving objects on a shelf, and face-washing using a washcloth—and outfitted them with wearable sensors to record motion capture data. Trained human coders watched the video recordings and marked the start and stop of motions, categorizing them using five classes: a “reach” for an object, a “reposition” to move closer to one, a “transport” to carry an object somewhere, a “stabilize” to keep an object still, or an “idle” to stand at the ready near one.
The team then trained the AI algorithms on these thousands of examples of motion and on the patterns of data representing each class—like a “reach” or a “transport”—until the program was able to count and classify the motions on its own. The resulting tool, which can process 6.4 hours of recorded activities in 1.4 hours versus the 513.6 hours human coders needed, brings a combination of pragmatism and precision the field has never seen before, Schambra said.
The field’s most common way to measure how much training a stroke patient receives during rehab is to go off of time—the hours the patient puts in during a given session. While practical, Schambra said, it’s imprecise. There’s incredible range in what patients could actually be doing in that time. “Say you’re going for an hour to the gym,” she said. “You might work out for 45 minutes with high-intensity cardio. I will probably sit in the sauna and have a smoothie. We could still be going to the gym for an hour apiece, but we’re doing much different physical quantities between the two of us.”
With the tool’s ability to measure training intensity both quickly and accurately, Schambra sees it as the ticket to finding out how many motions will ramp up recovery from stroke. She’s seen recent animal models that point to early, aggressive training with hundreds of repetitions of motions per session—10 times more than what patients are thought to be doing.
Schambra hopes to use the new approach in intensity response trials—similar to what you’d see in drug trials—in which researchers prescribe stroke patients (grouped by their level of impaired mobility) increasing quantities of movements, and study how well the patients recover based on the quantities prescribed. “We had to step back a notch and build this tool to be able to do these studies going forward,” Schambra said.
As a clearer picture of the motions made in rehab emerges and as researchers probe the quantity of movements stroke patients need to practice—especially in those crucial first weeks of peak neuroplasticity following a stroke—each patient may have a better shot at getting back to the basic abilities that let us move through life more smoothly than we realize.
The post Neurologists Used AI to Design a Better Stroke Recovery Program appeared first on The Daily Beast.