Wearable Sensor-Based Prediction Model of Timed UP and Go (TUG) Test in Older Adults
Advisor Information
Jong-Hoon Youn
Presentation Type
Oral Presentation
Start Date
26-3-2021 12:00 AM
End Date
26-3-2021 12:00 AM
Abstract
With the advancement of wireless sensor technologies, wearable sensors have been widely adapted to analyze human walking abilities. Although there has been a growing interest in using wearable sensor-based walking analysis in real-world conditions, most existing studies have been conducted in laboratory settings, limiting their applicability to capture and quantify walking patterns during activities of daily living. Thus, the purpose of this study is to develop a model that predicts timed up and go (TUG), which is one of the most commonly used measurements for detecting gait and balance problems. A total of 37 older adults were recruited for an experiment during which they wore wearable sensors to collect data on walking in real life. An elastic net and a ridge regression were used to reduce gait feature sets and build a predictive model. The predictive model reliably estimated the participants’ TUGs with a small margin of prediction errors (e.g., MAPE: 0.865). In addition, the best features and sensor locations for the prediction model were studied. We found that the foot sensor performed better than other sensors (e.g., MAPE: foot (0.865) > foot and pelvis (0.918) > pelvis (0.921)). We also found that whole-step- and double-stance-related features were the best set of features for predicting TUG score using the foot sensor. The proposed analysis model help clinicians effectively monitor patients’ fall risks remotely while allowing patients to go about their active lives.
Scheduling Link
1
Wearable Sensor-Based Prediction Model of Timed UP and Go (TUG) Test in Older Adults
With the advancement of wireless sensor technologies, wearable sensors have been widely adapted to analyze human walking abilities. Although there has been a growing interest in using wearable sensor-based walking analysis in real-world conditions, most existing studies have been conducted in laboratory settings, limiting their applicability to capture and quantify walking patterns during activities of daily living. Thus, the purpose of this study is to develop a model that predicts timed up and go (TUG), which is one of the most commonly used measurements for detecting gait and balance problems. A total of 37 older adults were recruited for an experiment during which they wore wearable sensors to collect data on walking in real life. An elastic net and a ridge regression were used to reduce gait feature sets and build a predictive model. The predictive model reliably estimated the participants’ TUGs with a small margin of prediction errors (e.g., MAPE: 0.865). In addition, the best features and sensor locations for the prediction model were studied. We found that the foot sensor performed better than other sensors (e.g., MAPE: foot (0.865) > foot and pelvis (0.918) > pelvis (0.921)). We also found that whole-step- and double-stance-related features were the best set of features for predicting TUG score using the foot sensor. The proposed analysis model help clinicians effectively monitor patients’ fall risks remotely while allowing patients to go about their active lives.