This is only useful for high functioning patients, for me it would recognize no use of the arm. Can't extend it, can't open hand, can't lift my arm above shoulder height unless I'm lying flat on my back.
A New Tool Helps Assess Rehabilitation Needed After Stroke
By Mary Beth Nierengarten
August 4, 2022
Article In Brief
Investigators were able to develop an artificial intelligence tool to quantify and recognize movements that could elicit the best outcomes for patients undergoing rehabilitation after stroke. Independent experts say it could help provide insights into how much training is necessary to optimize recovery after a stroke.
A new artificial intelligence tool provides a more precise way to assess how much rehabilitation is needed after a stroke to optimize best outcomes, according to a new study published on June 16 in PLOS Digital Health.
Developed by investigators at NYU Langone Health, the tool, called the Primitive Sequencing (PrimSeq) pipeline, is comprised of a wearable sensor array that captures upper-body motions—considered important for stroke rehabilitation—which a machine-learning algorithm has been taught to recognize and count.
The device could potentially close a gap in care that until now has limited the ability of clinicians to quantify and assess stroke rehabilitation, said the study's senior author Heidi Schambra, MD, associate professor in the departments of neurology and rehabilitation medicine at the NYU Grossman School of Medicine.
Knowing more precisely how much training is needed, and how intense rehabilitation exercises after stroke should be, is critical to ensure that patients are given the best opportunity to gain back function, said Dr. Schambra.
Evidence from preclinical work in animal models of stroke suggests that intensive training during the early weeks after stroke confers good outcomes, she said, but up until now, little was available to assess the most effective training intensity for humans, because there was no way to track how many arm motions are performed during stroke rehabilitation.
“It's a missing piece of information for training that has kept our field from identifying the best clinical interventions to achieve the best possible recovery from stroke,” Dr. Schambra said. Along with helping to identify training intensities that patients need in order to gain the best response to rehabilitation, quantifying and assessing how much rehabilitation after stroke is needed is also important for assessing the effects of adjunctive treatments like pharmacologic therapies and brain stimulation, she added.
“Most interventions are based on training, so if you are not accounting for how much each person is receiving, it is harder to interpret the effects of these adjunctive treatments on stroke rehabilitation,” Dr. Schambra said.
Study Details
To develop the tool, investigators modified a word recognition algorithm and trained it to recognize patterns of motion data captured from the wireless sensors. The algorithm was trained to identify patterns based on what Dr. Schambra described as building block movements for stroke rehabilitation— five classes of upper-body motions, including reach (a motion to make contact with an object), reposition (a motion to move into proximity to an object), transport (a motion to convey an object in space), stabilization (a minimal motion to hold an object still), and idle (a minimum motion to stand at the ready near an object).
Investigators trained the algorithm using motion data from 33 stroke patients—each of whom wore nine sensors—to glean in detail the motions occurring in the upper body. For example, data from the sensors would capture if the arms were moving in a linear direction or in a rotating manner. The device recorded more than 51,616 upper body movements, and each digital record of an arm movement was matched to one of the five classes of building block movements.
Once the algorithm was trained to recognize and quantify the motion data and motion patterns in the training group, investigators tested the algorithm in eight stroke patients that the algorithm had never encountered to see how the algorithm would work in a real-world situation. It accurately assessed most of the movements in patients who had mild-to-moderate arm impairments from stroke.
“It did really well in terms of counting the movements that patients were making, and also it did a solid job at correctly identifying the type of movement and classifying it correctly,” Dr. Schambra said.
Overall, she said the algorithm was about 77 percent accurate in identifying a motion; for example, if a patient did a “reach” motion the algorithm correctly identified it as a “reach” motion 77 percent of the time.
Dr. Schambra hopes to improve the tool's accuracy by teaching the algorithm to recognize more motions from stroke patients to include a wider reach of impairments.
The investigators also hope to make the device more user-friendly for clinicians. “Ideally we want to package this in a software program where clinicians can see how their patients are doing, and patients can see how they are doing, based on the number of motions and get feedback in real time,” she said, adding that this will allow clinicians to plan stroke rehabilitation more objectively for their patients. She emphasized that there is no patent on the tool, and that she and her colleagues aim to distribute the tool widely so that everyone has access to it.
Expert Commentary
David J. Lin, MD, a neurointensivist at Massachusetts General Hospital and director of the neurorecovery clinic, thinks the tool may be a game changer. “The approach has potential and could be paradigm-changing for stroke rehabilitation,” said Dr. Lin, who is also an assistant professor of neurology at Harvard Medical School.
Dr. Lin praised the investigators for specifically examining the feasibility and practicality of the tool in clinical settings by examining patients' ratings of the time required for setup and potential discomfort experienced by wearing the sensors. For both, the experiences of patients were overall very positive with minimal burden, said Dr. Lin. Dr. Lin thinks it would be good in the future to survey therapists about the feasibility of incorporating the sensors into their clinical practice.
Dr. Lin also would like to see further examination into the fundamental assumptions on which the algorithm is built, namely, the five classes deemed to be the building blocks of functional movement used in the study. For example, he'd like to see other potential features examined, such as looking into whether there are ways to quantify functional movements without making prior assumptions. “Could there be automated (i.e., computer vision) techniques for identifying the building blocks of functional movement?” he asked. He also questioned how these building blocks relate to patterns of central nervous system activation and their disruption after stroke.
No comments:
Post a Comment