Subjects completed tasks typically performed during in-person clinical assessments by collecting data with sensors attached to their arms, hands, and feet.Credit: Manuel Enrique Hernandez
× close
Subjects completed tasks typically performed during in-person clinical assessments by collecting data with sensors attached to their arms, hands, and feet.Credit: Manuel Enrique Hernandez
Low-cost wearable sensors could increase access to care for people with Parkinson's disease. Researchers and clinical collaborators at the University of Illinois at Urbana-Champaign show in a new study that a new machine learning approach and a baseline of data from healthy older adults improve the accuracy of results from these sensors. I discovered it inThe results were published in a magazine sensor.
Study leader Manuel Enrique said: “This study shows that the expansion of the dataset to include movement data from healthy older adults and its integration with deep learning approaches will help patients with Parkinson's disease for use in future telemedicine sessions. “This study shows that it can help improve the accuracy of detecting differences in motor dysfunction.” Mr. Hernandez is the Carl Professor of Biomedical and Translational Sciences at the University of Illinois College of Medicine.
To monitor symptoms, Parkinson's patients need regular evaluations, which usually require in-person visits with specialists that are time-consuming and have limited availability, Hernandez said.
Telemedicine assessments would improve patient access to care, but are hampered by the lack of quantifiable measurements. For example, a doctor can see a patient's movements, but cannot assess stiffness or muscle tone. Although some progress has been made in the use of wearable sensors to assess athletic performance, cost limits their use.
To address these challenges, the Illinois team focused on how to improve assessments using low-cost wearable sensors.
“Ideally, we would have something that is completely passive, collecting data as a person moves through their daily environment, and using that information to provide guidance on overall motor function and the progression of neurological symptoms.” “We would like to do that. But then we face a big challenge: how do we do it? Parse all the information to make it useful to clinicians,” Hernandez said.
“This led to our strategy to improve machine learning, focus on activities that are useful for assessment, and look at healthy people as a baseline.”
The researchers adopted a revised version of the Movement Disorders Society-sponsored Unified Parkinson's Disease Rating Scale, the gold standard for clinical assessment. MDS-UPDRS outlines specific tasks performed by patients and qualitative observations made by physicians during examinations and organizes them into categories for scoring.
In the study, researchers had patients perform tasks and muscle movements while wearing the sensor to provide data for the categories scored by MDS-UPDRS. They trained a machine learning model using data from both Parkinson's disease patients and healthy older adults.
“Improving machine learning models requires large amounts of data. We hypothesized that healthy people could serve as a basis for predicting potential age-related changes. . We can then build comparative models to understand when significant changes occur, either in terms of motor function or motor impairment,” Hernandez said.
The researchers also used a deep learning technique known as pre-training to make the model more robust, allowing it to better identify important data points and filter out irrelevant data points. With pre-training and the addition of data from healthy older adults, he improved his accuracy in identifying motor impairments in hand-moving tasks by more than 12% over current standard models. Additionally, including data from healthy older adults improved the accuracy of assessing all upper and lower body tasks except toe tapping.
Next, the researchers hope to expand the model by further collaborating with neurologists and clinicians. They aim to identify a small number of additional tasks that can quantitatively measure more information about classic symptoms of Parkinson's disease, such as tremor, while maintaining the low cost and ease of use of the sensor.
“There is a great need to better understand and better quantify the ongoing changes in Parkinson's disease,” Hernandez said.
“The ability to use wearable sensors while performing activities that are part of a clinical assessment in a telemedicine environment provides a more objective and useful way to directly support and, hopefully, improve the quality of care for patients with Parkinson's disease. It can open the door to quantifiable information. Life moves forward.”
For more information:
Mehar Singh et al. A deep learning approach for automatic and objective grading of motor deficit severity in Parkinson's disease for use in remote assessment, sensor (2023). DOI: 10.3390/s23219004
