Parkinson's Tremor Onset Detection and Active Tremor Classification Using a Multilayer Perceptron

Abstract

The study of the characteristics and behaviour of tremor for people suffering from Parkinson's disease (PD) is an important first step in developing a new method to predict future tremor signals, their onset and the active tremor instances. The current approaches to detect tremor are limited to tremor estimators that rely on simple tremor models, or on deep brain probing that is invasive in nature. Thus, a new method that is noninvasive and that can capture tremor complexity to predict when tremor is active is needed. In this work, a new approach is presented using neural networks (NNs) and data from inertial measurement units (IMUs) to predict tremor onset and classify the active tremor instances in the wrist and metacarpophalangeal (MCP) joints of the index finger and thumb. The developed model showed an accuracy of 92.9% in predicting and detecting tremor onset, and therefore can be considered a reliable tool that has the potential to be integrated with wearable assistive devices for suppressing tremor.