Tremor quantification of Parkinson's disease - a pilot study

摘要

The objective of this study was to quantify Parkinson's disease tremor severity using signals acquired from wearable inertial sensors. A machine learning approach was used in the development of classification models. Features calculated from signals produced by accelerometer and gyroscope sensors placed on the index finger and wrist were taken into account. Linear Support Vector Machine (SVM) models were used to assess the severity of rest and postural tremor. The analysis showed that sensor signals collected from the index finger more accurately predict tremor severity compared to signals from a sensor placed on the wrist. Also, standard deviation of linear acceleration and angular velocity signals was shown to increase the accuracy of classifier from 88.6% to 88.9% for resting tremor and 78.8% to 81.8% for postural tremor.