Researchers at Florida Atlantic University have developed a new method using wearable sensors and AI that could revolutionize the practice of balance assessment. Photo by Alex Dolce, Florida Atlantic University.
Traditionally, clinicians have relied on subjective observation and specialized equipment to measure balance in people with conditions like Parkinson's disease, nerve injuries, and age-related decline. Such methods, especially subjective methods, can lack precision, be difficult to administer remotely, and lack consistency. To address these limitations, researchers at Florida Atlantic University have developed a new approach using wearable sensors and advanced machine learning algorithms that could redefine the practice of balance assessment.
The study is published in Frontiers in Digital Health.
Sensor setup
The researchers used wearable inertial measurement unit (IMU) sensors attached to five locations: the ankle, lumbar spine, sternum, wrist, and arm. Data collection followed the modified Clinical Trial of Sensory Interaction in Balance (m-CTSIB) protocol, testing four sensory states, eyes open and closed, on stable and foam surfaces. Each test lasted approximately 11 seconds, simulating a continuous balance scenario.
The scientists then preprocessed the raw sensor data to extract features and applied three machine learning algorithms — multiple linear regression, support vector regression, and the open-source software library XGBOOST — to estimate the m-CTSIB score.
AI Balance Detective Training
The researchers trained and validated their model using wearable sensor data as input and the corresponding m-CTSIB scores from Falltrak II as ground truth labels.
They evaluated the performance using cross-validation methods, correlation with ground truth scores, and the mean absolute error (MAE) measure.
The XGBOOST model using lumbar sensor data produced the best results, demonstrating high accuracy and strong correlation with the actual balance score. The lumbar and dominant ankle sensors performed best in estimating the balance score.
Towards a more accurate balance assessment
“Our findings pave the way for more accurate and convenient balance assessment,” the researchers conclude in Frontiers in Digital Health. They say their approach “has great potential to enhance the assessment and management of balance performance in a variety of settings, including clinical settings, rehabilitation, and remote monitoring.”
