Highlight
Automatic identification of motor states in people with Parkinson's disease
Machine learning of tremor from digital pen test
Achievement/Results
Parkinson’s disease (PD) is a progressive neurodegenerative disorder that causes fluctuating motor deficits such as akinesia, bradykinesia, impaired balance, and tremor. Time periods characterized by severe deficits are referred to as “off” periods, while periods of relatively normal function are considered “on” periods. Clinicians treat these deficits through the combined administration of Carbidopa/Levodopa medication, dopamine medication, and deep brain stimulation. Overmedication or overstimulation can cause dyskinesia, an involuntary, often rhythmic or choric, exaggeration of movements. Clinicians assess the occurrence of off periods and dyskinesia, in part, by asking patients to retrospectively self-report how frequent these periods occurred over the several months prior to the clinical visit, a method subject to recall bias. To augment this assessment, several systems have been developed to provide clinical ratings from body-worn sensor data using computational algorithms. However, these systems place a substantial time burden and inconvenience to both clinicians and patients. Our ultimate goal is to develop an objective system that continuously identifies on, off, dyskinesia, and non-dyskinesia periods occurring over a 7-day period without placing a time burden on a clinician.
We hypothesize that we can classify body-worn accelerometer data into on, off, dyskinesia, and non-dyskinesia periods using signal analysis, feature extraction, and machine learning algorithms (MLAs) combined with ecological momentary assessment (EMA). To achieve our goal, IGERT trainee Nathan Darnall first completed a study in which gyroscopic data was automatically classified onto tremor severity ratings with 82% accuracy during clinical assessment tasks. He next wanted to classify data collected with 5 body-worn triaxial accelerometers during activities of daily living (ADL) onto dyskinesia periods. Preliminary results showed 96% accuracy using a decision tree MLA. Currently, he is collecting accelerometer data combined with EMA reports for a 7-day study with the goal of classifying data onto, off, dyskinesia, and nondyskinesia periods using computational methods similar to the previous studies. He will validate the ability of this system to automatically and continuously generate a time history of on and off, dyskinesia, and non-dyskinesia periods, which clinicians could use to optimize their prescription of PD treatments.
Address Goals
This particular health technology challenge was posed to our IGERT Gerontechnology class by Jonathan Carlson, a neurosurgeon who performs deep brain stimulation in Spokane, Washington. IGERT trainee, Nathan Darnall, a mechanical engineering student at Washington State University, performed an internship with Dr. Carlson in order to better understand current assessment techniques and to observe deep brain stimulation procedures. He then designed an approach to measure dykinesia using wearable sensors and machine learning algorithms. This research project represents a multi-disciplinary challenge that was made possible only because of the IGERT support for new multi-disciplinary classes, for visitors such as Dr. Carlson, and for internships that allow STEM students to become familiar with the medical challenges that they are trying to address.
