Falls Detection, Prediction and Cognitive Decline Assessment Based on Inertial Sensors

Abstract

The elderly population is growing rapidly worldwide and is expected to reach 1.6 billion by 2050. Cognitive impairments or dementia are quite prevalent among elderly people and their incidence rate increase with age. There are numerous medical conditions that can cause dementia such as neurodegenerative diseases, vascular disorders, excessive use of alcohol, to name a few. A direct consequence of cognitive decline is behavioral changes such as disturbances in speech, ADL's, and gait patterns. This cognitive decline alone or sometimes in conjunction with other medical problems (musculoskeletal impairments, sensory system issues) leads to mobility disorders among elderly people. These movement disorders such as gait and balance impairments ultimately cause fall events. The consequences of falls are very severe and long lasting that include injuries, disability, long-term hospitalization, reduced mobility and self confidence, fear of falling - which further results in social isolation and decreased quality of life, and even death. In order to enhance quality of life and healthy life expectancy of elderly people, the early detection of those with cognitive impairments and high falls tendency is extremely important (preferably under living conditions), so that the reversible causative factors can be identified and treated.

The techniques for human body motion analysis can be broadly classified into two categories: 1) context-aware systems and 2) wearable systems. The former category concerns the deployment of sensory equipment such as cameras, ultra-wideband radar, pressure sensitive platform etc. and offers an unobtrusive solution. Such systems are mostly limited to clinical settings due to their high cost, limited coverage, and privacy concerns (video-based systems). Wearable motion sensors that rely on kinematic signals, like triaxial accelerometer and gyroscopes, fall under the latter category. The advancement in microelectromechanical systems has enabled the development of compact, low power, lightweight and inexpensive wireless inertial sensors that eliminate the limitations of context-aware systems. Inertial sensors signals can be used to extract numerous movement-specic quantitative features in order to measure and analyze the functional mobility, balance and gait of the wearer. The work presented in this dissertation, mainly focuses on inertial sensors based objective and pervasive assessment of functional mobility, gait, and balance of elderly people, for the purpose of realtime falls detection, screening of elderly people with high fall-risk and (mild) cognitive impairments in their home settings.

Common fall occurrences in the elderly population pose dramatic challenges in public health care domain. Adoption of an efficient yet highly reliable automatic fall detection system may not only mitigate the adverse effects of falls through immediate medical assistance, but also profoundly improve the functional ability and confidence level of elder people. Successful deployment of a fall detection system among elderly population depends on various factors: usability (number of sensors, their positions, prefix orientation, battery lifetime etc.), privacy issues, cost, and reliability (accuracy, false alarm rate). As the existing literature related to these issues is lacking, there is no widely accepted fall detection system (FDS) among elderly until now. Hence, it is of great significance to develop a reliable FDS to fill this gap while accounting for the existing practical issues. We present a pervasive FDS developed on smart phones (SPs) namely, FallDroid that exploits a two-step algorithm proposed to monitor and detect fall events using the embedded accelerometer signals. The proposed algorithm uses novel techniques to effectively identify fall-like events and reduce false alarms. In addition to user convenience and low power consumption, experimental results reveal that the system detects falls with high accuracy and achieves the lowest false alarm rate till date.

In addition to fall event detection and emergency alert system, it is necessary to identify high fall-risk population in advance and to prevent falls by early intervention. In order to identify fallers promptly and accurately, numerous functional mobility assessments have been suggested. Recently, many sensor assisted mobility assessments have achieved better classication results as compared to their original versions that are mostly subjective in nature. The existing literature on the wearable inertial sensor based fall risk assessment techniques are predominantly grounded on binary classification problem i.e. screening of fallers and non-fallers. To date, only few works have focused on regression problem and tried to replicate the (continuous scale) score of existing validated clinical fall-risk assessment tools. However, most of the existing works present over-optimistic and highly questionable results mainly due to improper modeling decisions, problematic validation and features selection approaches. While addressing the problems of existing works, we proposed a Berg Balance Scale (BBS) score estimation model using trunk acceleration based features, where BBS is a widely used validated clinical test to assess person's balance abilities and is also an effective measure for fall-risk prediction. Our results demonstrate that the proposed method has the potential to act as a surrogate of standard clinical assessment-BBS and to assess quantitative fall-risk in an unsupervised setting. Furthermore, we explored the comparative ability of the three prevalent sensor assisted mobility assessments and their novel combinations for classication of fallers.

The incidence and prevalence of Mild Cognitive Impairment (MCI) is high among older people. MCI is a transitional cognitive state between normal cognition and dementia. For early diagnosis of dementia, MCI screening and early diagnosis is extremely important preferably under living conditions. As gait is a complex cognitive task, instrumented gait assessment may allow the detection of subtle motor deficits related to MCI, allowing early detection and intervention of cognitive impairment. To date, only few studies reported the instrumented gait assessment of MCI people. Most of them utilized pressure sensitive platform (GAITRite) in a clinical setting to evaluate and report standard spatio-temporal gait parameters. The literature on inertial sensor based comprehensive gait analysis of MCI people is lacking. To fill this gap, we analyzed the gait of MCI and Cognitively Normal (CN) subjects under single and dual-task walking conditions using shank mounted inertial sensors. We reported many gait biomarkers that can assist MCI diagnosis and facilitate the early detection of dementia. Further more, we proposed a Computer Aided Diagnosis (CAD) system for automated prescreening of CN and MCI people using inertial sensors based pervasive gait analysis under dual-task walking scenario.