Date of Award

Spring 2018

Document Type



Santa Clara : Santa Clara University, 2018.


Computer Engineering; Bioengineering

First Advisor

Yuling Yan

Second Advisor

Behnam Dezfouli


Parkinson’s Disease (PD) is a progressive neurological disease that affects 6.2 million people worldwide. The most popular clinical method to measure PD tremor severity is a standardized test called the Unified Parkinson’s Disease Rating Scale (UPDRS), which is performed subjectively by a medical professional. Due to infrequent checkups and human error introduced into the process, treatment is not optimally adjusted for PD patients. According to a recent review there are two devices recommended to objectively quantify PD symptom severity. Both devices record a patient’s tremors using inertial measurement units (IMUs). One is not currently available for over the counter purchases, as they are currently undergoing clinical trials. It has also been used in studies to evaluate to UPDRS scoring in home environments using an Android application to drive the tests. The other is an accessible product used by researchers to design home monitoring systems for PD tremors at home. Unfortunately, this product includes only the sensor and requires technical expertise and resources to set up the system. In this paper, we propose a low-cost and energy-efficient hybrid system that monitors a patient’s daily actions to quantify hand and finger tremors based on relevant UPDRS tests using IMUs and surface Electromyography (sEMG). This device can operate in a home or hospital environment and reduces the cost of evaluating UPDRS scores from both patient and the clinician’s perspectives. The system consists of a wearable device that collects data and wirelessly communicates with a local server that performs data analysis. The system does not require any choreographed actions so that there is no need for the user to follow any unwieldy peripheral. In order to avoid frequent battery replacement, we employ a very low-power wireless technology and optimize the software for energy efficiency. Each collected signal is filtered for motion classification, where the system determines what analysis methods best fit with each period of signals. The corresponding UPDRS algorithms are then used to analyze the signals and give a score to the patient. We explore six different machine learning algorithms to classify a patient’s actions into appropriate UPDRS tests. To verify the platform’s usability, we conducted several tests. We measured the accuracy of our main sensors by comparing them with a medically approved industry device. The our device and the industry device show similarities in measurements with errors acceptable for the large difference in cost. We tested the lifetime of the device to be 15.16 hours minimum assuming the device is constantly on. Our filters work reliably, demonstrating a high level of similarity to the expected data. Finally, the device is run through and end-to-end sequence, where we demonstrate that the platform can collect data and produce a score estimate for the medical professionals.