Advancing Clinical Gait Assessment Methods with Low-Cost Triaxial Accelerometers: Applications for Individuals with Neurodegenerative Diseases

Abstract

Clinical gait assessment provides objective measurement of a person’s locomotor and dynamic balance control. Applications of wearable sensors for gait assessment have gained popularity over the past decade as a feasible tool to objectively measure gait in clinical settings. Commonly, clinical assessments and research studies limit a focus to steady-state walking relying on static thresholds to remove intra-bout phases, such as gait initiation and termination. Despite this historical focus on steady-state gait, the use of wearable technology affords the ability to provide insight into the variability of walking that occurs not only within (intra-stride) and between strides (stride-to-stride) but also variability of phases within a walking bout (intra-bout), measures that potentially have clinical importance. To achieve necessary details of intra-stride, stride-stride and intra-bout characteristics there continues to be a need to advance algorithm development to accurately detect gait characteristics specifically when applying across range of ages, task conditions, and gait control ability. The purpose of this thesis was to advance the analytical approaches of gait assessments performed in clinical environments with cost-effective wearable sensors in two meaningful ways: (1) advancing stride segmentation methods beyond standard stride or stance-swing segmentation using ankle-worn accelerometry and describe these metrics across tasks of varying difficulty, (2) identifying groups of strides that are representative of intra-bout phases and how these metrics change with task and understand their relation to gait-specific clinical metrics of mobility in older adults and neurodegenerative disease. Using accelerometry data collected as part of the Ontario Neurodegenerative Disease Research Initiative (ONDRI) allowed for the development of a Finite State Machine (FSM) algorithm to segment gait cycles into unique intra-stride phases. The FSM algorithm was tested for stride detection accuracy, across cognitively intact young adults and people living with neurodegenerative disease (NDD) or cerebrovascular disease (CbVD), and during walking with varying speeds and secondary cognitive tasks. Temporal and accelerometry-based kinematic outcomes from the algorithm were evaluated across dual-task walking conditions to identify compensatory gait strategies to secondary cognitive demands. These same FSM outcomes were used to quantify intra-bout phases using stride clustering approaches and examine how intra-bout phases and accelerometry-based kinematics might contribute to stride-to-stride variability within the dual task paradigm. In addition, the relationship between conventionally reported metrics, stride clustering outcomes, and gait-specific clinical markers was assessed to understand the interaction between mobility outcomes. In general, FSM stride detection performed well across tasks of varying difficulty and speed in young healthy adults during long treadmill walking and in people with NDD or CbVD during short overground walking tasks. Algorithm performance was also considered in the context of defining long periods of walking for implications in free-living environment; where depending on the parameters used to define a walking bout, the algorithm performance influenced the identification of the start and stop of walking. This approach to segment walking into stride-to-stride and intra-stride phases proves to be an accurate method using minimal data inputs with clinically feasible tools. Temporal and accelerometry-based kinematics (e.g., cycle time, accelerometer magnitudes) derived from the FSM methods were different between clinical walking tasks, such that cycle time gradually increased during dual task walking and across task difficulties and accelerometry kinematics decreased during dual task walking. Of particular interest was the significant differences between preferred walking tasks; the first preferred trial was typically different from the subsequent trials with the same conditions for conventionally reported and accelerometry based gait outcomes. In addition, when these outcomes were used to characterize intra-bout phases there were significant differences between preferred and dual task conditions across all cluster-based outcomes (e.g., initiation length). These outcomes suggest that the FSM can detect previously reported and novel kinematic outcomes to describe compensatory strategies selected by NDD and CbVD persons in response to secondary dual task walking. Clinically, these FSM or cluster-based outcomes can distinguish between groups that are stratified by commonly reported gait-specific markers of mobility, such as walking speed, dual task cost magnitude, or the Unified Parkinson’s Disease Rating Scale gait score. This work advances the detection of stepping with an adaptable algorithm that requires minimal data input, performs well across young and older adults including those with NDD, and can characterize compensatory strategies under dual task conditions. In addition to algorithm advancements, the present thesis expands stride clustering methodology to identify and characterize intra-bout phases across different dual-task conditions and understand how these representations of stride variability are related to clinical gait outcomes. Results from this study highlight the need to examine the contribution of intra-bout phases to gait variability and emphasizes the opportunity to expand our analytical approaches to gait assessment in and outside the clinic for optimal rehabilitative and therapeutic interventions.