Deniz Erdogmus, Stephen Intille. Paolo Bonato
Date of Award
Doctor of Philosophy
Department or Academic Unit
College of Engineering, Department of Electrical and Computer Engineering
electrical engineering, biomedical engineering, COPD, machine learning, Parkinson's disease, remote health monitoring, stroke, wearable sensors
Biomedical | Electrical and Computer Engineering
The focus of this dissertation is on the application of wearable sensor technology for objective monitoring of motor symptoms in patients with mobility limiting conditions. Recent developments in the field of wearable sensors have enabled long-term health monitoring in a variety of patient populations. A large amount of raw sensor data is gathered in a typical long term monitoring scenario. Hence, it becomes extremely important to develop data analysis methods to extract clinically relevant information from raw sensor data and present it to clinicians in an easily readable format. The main aim of this work was to develop methods for extracting meaningful clinical information from accelerometer data recorded during the performance of motor activities. Three clinical applications were explored in this context.
Quantitative assessment of motor abilities in stroke survivors can provide valuable feedback to guide clinical interventions. The Functional Ability Scale (FAS) is a 75-point scale used to evaluate the upper extremity functional ability of subjects by grading movement quality during performance of 15 motor tasks. We developed a methodology to derive estimates of FAS clinical scores from accelerometer data recorded during the performance of functional motor tasks. We have shown that a Random Forest classifier trained using features extracted from accelerometer data can accurately estimate the clinical scores associated with individual motor tasks. In addition, we trained a simple linear regression model for estimating the total FAS score based on scores derived from individual motor tasks. We also performed an analysis to determine a small set of sensor locations and motor tasks required to accurately estimate the total FAS score.
In Chronic Obstructive Pulmonary Disease (COPD), exacerbations have been linked with higher mortality rates and a reduction in health-related quality of life. Early intervention can lead to better outcomes. It has been shown that during an exacerbation, physical activity level is significantly reduced. We have developed a methodology for activity recognition based on features extracted from accelerometer data recorded during the performance of physical activities. In our study, subjects performed motor activities of daily living as well as exercises typically prescribed during pulmonary rehabilitation. By exploiting the hierarchical relationships between clusters belonging to different activities, we developed a two stage classification method using a Support Vector Machine (SVM) classifier. Our method showed an improvement in the classification accuracy of activities, which are similar biomechanically but different from the systemic point of view. We also studied the impact of number of sensors locations on the classification accuracy as well as the computational costs associated with extracting data features directly on the sensor nodes. We have proposed an optimized system for long-term activity monitoring of COPD patients in the home setting.
In patients with Parkinson's disease (PD), regular monitoring of the severity of motor symptoms such as tremor, bradykinesia and dyskinesia plays an important role in the clinician's choice of therapeutic interventions for disease management. Currently this is achieved via patient self reports and motor assessments performed by clinicians in the hospital setting. Both these approaches are limited due to their subjective nature. In addition, due to the time consuming nature of clinical assessments, they are performed infrequently. In three separate studies, we have gathered data using accelerometers from PD patients' during the performance of motor tasks. We showed that features extracted from accelerometer data can be used not only to capture information about symptom severity but also to understand the rate of change of different motor symptoms. We implemented and optimized a Support Vector Machine (SVM) classifier for estimating the severity of tremor, dyskinesia and bradykinesia. We have also analyzed the importance of different feature types on the classification accuracy and the cost of estimating these features directly on the sensor nodes. We have further extended this analysis by implementing a regression based SVM to longitudinally estimate the severity scores for two motor tasks.
Patel, Shyamal, "Quantitative motor assessment in patients with mobility limiting conditions using wearable sensors" (2012). Electrical Engineering Dissertations. Paper 45. http://hdl.handle.net/2047/d20002546
Click button above to open, or right-click to save.