A Wearable Health Monitor To Aid Parkinson Disease Treatment

This thesis developed a wearable motion capture system to record Parkinson's patients performing daily activities during each of the three stages of their medication cycle. Five calibrated accelerometers continuously monitored motions of the subjects' torso, wrists and ankles, and stored the resulting data onto a low-cost compact flash (CF) memory card. Five hours of data was recorded from a volunteer with PD wearing the motion recording system, along with the corresponding medication state rated by an observing physician. This data was divided into training and test sets, where one-quarter was reserved for testing. A neural network demonstrated 85% correlation between data sampled from all five accelerometers to the dyskinetic medication state labelled by the physician. Noting inherent confusion between a sedentary patient with high dyskinesia and a properly medicated patient moving energetically, a second neural network was trained to identify periods of walking, with 75% correlation. Using this activity classifier to remove periods of walking increased the overall accuracy to 91%. Thesis Supervisor: Alex "Sandy" Pentland Title: Professor of Media Arts and Science A Wearable Health Monitor To Aid Parkinson Disease Treatment