Session: 07-08-01: Biomedical Devices, Sensors, and Actuators I

Paper Number: 165161

Fall Risk Analysis With a Machine Learning Model on Smartphone-Collected Motion Signals 

Falls carry significant risks in older adulthood, including serious injury and death. It is imperative to assess and implement efficient countermeasures to limit the number of falls, as well as their effects, with the growing older adult population. There are numerous reasons for the increased risk of falling, ranging from physical impairment to external factors such as uneven and slippery floor surfaces. Understanding these causative conditions will help healthcare professionals develop more successful prevention strategies and improve the overall health of older adults. One of the most well-known fall risk assessments is the Timed Up and Go (TUG) test. This clinical test measures how long it takes an individual to stand up from a chair, walk 3 meters, turn, and sit down again. The TUG duration assesses a patient's risk of falling by evaluating balance and stability while walking. Similarly, the Berg Balance Scale (BBS) is another clinical test that quantifies an individual's balance. It is composed of activities that measure balance performance, and the greater the score, the better the balance. The Berg test is an extremely reliable method for assessing one's capacity for stability when on the move. Both the TUG and BBS are effective in assessing fall risk and measuring one's physical capacity at the time. Likewise, the Five Times Sit to Stand (5TSTS) test and 6-minute walk test were employed as they proved to be reliable and invaluable balance measures. In the 5TSTS test, subjects were required to repeatedly stand up and sit down, while the 6-minute walk test measures the distance walked during this time. While the above-mentioned tests need to be evaluated by professionals in a clinical or research facility, motion sensors embedded in wearable devices and smartphones, like accelerometers and gyroscopes, could be used to capture these measurements. There is also discussion on whether the pattern of the motion signals could be analyzed to further quantify fall risk. Previous research has offered models for assessing everyday motion signals, although with varied success. This approach has been widely examined as the use of smartphones is pervasive and sensors are already built in modern smartphones. Our study differs by implying a combination of traditional test protocols, such as TUG and Berg, and modern sensor-based techniques. A mobile application has been built to collect comprehensive motion data in a smartphone with Android software and gold-standard IMU sensors were used as benchmarks to check signal accuracy. By analyzing data collected by both consumer and research-grade sensors, the study aims to enhance the precision of fall risk evaluation using smartphones. Machine learning models, such as linear regression, trees, and SVM (Support Vector Machines), etc. were tested using smartphone data to estimate the BBS, an optimal model is introduced in the paper with a discussion of the results. The technique provides a more stable way to determine fall risk among individuals such as that timely interventions may be offered to prevent falls and promote safety among older adults.

Presenting Author: Linda Zhu University of Michigan - Flint

Presenting Author Biography: Dr. Linda Zhu is currently an associate professor of mechanical engineering at the University of Michigan - Flint. Dr. Zhu’s current research focuses on NVH (Noise, vibration, and harshness), signal processing and sensor technologies, bioengineering, and neurological disease diagnosis. Her overall research goal is to develop innovative technologies that solve challenging problems in medical science, mechanical, and biomechanical engineering. Her current research projects include:
• Sound source localization, de-noising, and structural health monitoring
• Neural computation
• Neurological disease diagnosis and medical devices development

Authors:

Fardeen Mazumder University of Michigan - Flint
Abuelgasim Mohamed University of Michigan - Flint
Linda Zhu University of Michigan - Flint
Cathy Larson University of Michigan - Flint
Jennifer Liao University of Michigan - Flint
Charlotte Tang University of Michigan - Flint
Nathaniel Miller University of Michigan - Flint