Abstract
The developments in signal processing, data mining, and machine learning tools, supported by the advent of body sensor networks, capable of collecting high-precision and continuous data in a non-invasive fashion, resulted in great interest in manipulating gait data into equivalent objective markers for identification of pathological gait or other functional and cognitive impairments, monitoring ailments, evaluating the efficacies of the treatments, guiding recovery and rehabilitation, sports training, and enabling self-monitoring. Gait analysis has become a crucial assessment tool in medicine. It is used to provide new insights to help understand various human movement patterns and fluctuations in them corresponding to pathologies and neurological conditions affecting motor and/or cognitive functions. Multiple sclerosis (MS) is an example of such disorders. People with MS represent a heterogeneous cohort with a broad spectrum of symptoms. Gait dysfunction is a common finding, but one with varied etiology.
Assessments of MS-associated cognitive and motor disability, the disease course and its progression, and decision-making regarding disease-modifying treatments and symptoms management, are based on clinical observations, comprised of outcomes of physical examinations and medical imaging, and patient-rated questionnaires. Being reliant on physicians' judgment in interpreting imaging and clinical outcomes, affected by the differences among individuals regarding the notion of disability or improvements, time-consuming, imprecise, having limited sensitivity to subtle changes in gait, and low variance in ratings are some of the drawbacks of current subjective evaluations. Moreover, patient-reported outcomes (PROs) are subject to response shifts due to changes within individuals over time regarding health standards, and could lead to confusing findings and discrepancies between expected and observed indicators, negatively impacting disease prognosis. We intend to augment existing information, on-going research, and currently-used speed and distance-based clinical assessments, for a neurologic condition, with new, objective, and clinically meaningful anchors. Although our goals are motivated for a target application (finding physiologically meaningful inertial measures for assessing functional quality in MS using inertial gait data), our test measures could be adopted for gait assessments and monitoring in other neurological disorders, balance, stability, and fall risk prediction, and general health and wellness applications.
We derive inertial features using rotational acceleration gait data collected using a body sensor node for improved gait assessment with three goals -- (i) using variations in gait features over time, i.e., gait dynamics, to remove the inter-subject variability and guide personalized assessments in neurology-affected locomotion, aging, or chronic diseases, (ii) finding computationally efficient and robust gait features that neither require identification of exact gait cycles nor need a large dataset to capture gait deterioration and make physical sense, (iii) translating pathology-induced fluctuations into frequency-domain features to identify the impact of MS on important gait phases. We include gait variables for comparable controls in the study to determine the unique and overlapping features.
We use a class of data mining to generate models estimating the relationships between the inertial measures and clinical assessments and identifying the physiological significance of the feature space to make it accessible to the doctors and allow patient's engagement. The relationships between the inertial and clinical measures also help to identify the clusters of gait variables, besides and beyond what are identified using a traditional disability assessment scale, to confirm whether the scale is the best way to categorize persons with MS into different disability levels based on gait impairment.
To conclude, we use inertial measures to verify three hypotheses -- (i) pathologic gait in MS is restricted and, thus, is less variable in comparison to a healthy gait, (ii) distinct types of MS disability introduces disturbance in various gait components, and (iii) changes in gait variables over time carry additional information about the disease status. We believe that knowledge of our test measures will improve our capacity to monitor the disease and its progression, evaluate the effectiveness of the treatments, improve and tailor subjective assessments based on individual needs, and guide self-management of symptoms.
