Falls are the number one cause of injury and injury-related deaths in older adults. Nearly one-half of those over age 65 are unable to rise independently after falling, and a significant source of morbidity is the ‘long-lie’ that often occurs after falling. A wearable sensor system that automatically detects falls can facilitate quicker delivery of care. Such systems can also log information on the nature of the fall to inform prevention efforts. This thesis describes my efforts to develop improved methods for detecting fall-related events in older adults through wearable sensors (i.e. accelerometer and gyroscopes). In particular, I developed and evaluated novel approaches to extend the utility of fall monitoring systems beyond post-impact fall detection, to pre-impact fall detection, near-fall detection and causes of fall detection. In my first study, I conducted laboratory experiments to compare the accuracy of machine learning versus thresholdbased approaches for distinguishing falls from daily activities based on wearable sensor data. In my second study, I examined the accuracy of machine learning algorithms in distinguishing falls from real-world fall and non-fall datasets from young and older adults. My third study focused on pre-impact fall detection (detecting falls during the descent phase before impact) which is relevant to the design of active protective gear (e.g., airbags). In particular, I determined how the data window size and lead-time affects classification accuracy based on a single waist sensor. In my fourth study, I developed a near-fall identification algorithm based on machine learning, which could provide biofeedback to the individual of their state of balance. I examined how the number and location of sensors on the body influenced the accuracy of the algorithm in identifying near-fall from activities of daily living. My final study examined the ability of wearable sensors to provide objective evidence on the cause and circumstances of falls, to aid in diagnosing and treating the underlying causes of falls in older adults. My overall efforts advance the potential of wearable sensors (i.e. accelerometers and gyroscopes) for providing objective and clinically relevant information for the prevention and treatment of falls and their related injuries in older adults.
Keywords:
older adults; falls; wearable sensors; machine learning; biomechanics; biomedical engineering