Differentiation of Parkinson’s Tremor from Essential Tremor using Action Tremor Analysis

Abstract

Similarities in the physiological and psychological symptoms of Parkinson’s disease (PD) and Essential Tremor (ET) make accurate diagnosis of PD and ET conditions difficult. It has up to 25% diagnostic error rate. Both disorders have similar postural tremor characteristics, which make it difficult to differentiate on the basis of tremor between the two disorders. Previous studies that have classified PD and ET tremor used multiple neural learning tools and decision support systems to classify between the two conditions. In contrast, we explored the use of Discrete Wavelet Transforms combined with Support Vector Machines and changes in cognitive load as a disorder classification method.

Advancements in mobile device technology allow these devices to be used as mobile medical applications. This provides the opportunity for one device to collect, analyse and classify biometric data from a range of disorders. Biometric data has been collected by mobile devices for a number of years, but most analysis and classification have been performed off-line on a central server.

In this research, we have concentrated on two questions: Can changes in tremor be used to classify PD and ET postural tremor on a mobile phone? Do the effects of changes in cognitive load through differing levels of attention and distraction affect the level and frequency of tremor differently in PD and ET disorders? We investigate the influence of attention and distraction on tremor by comparingParkinson’s postural tremor with Essential postural tremor as the between-subjects factor and differing attention and distraction levels as within-subjects factors. A primary finding is that in the frequency band directly related to postural tremor in both Essential Tremor and Parkinson’s Disease there is significant difference in the disorders between attention and distraction tasks. These findings suggest that attention and distraction can be successfully used as an input feature space to classify difference in these two disorders. Using the differences between attention and distraction tasks we successfully developed a proof-of-concept smartphone based mobile medical application using discrete wavelet transforms and support vector machine based classification to discriminate between Parkinson’s postural tremor and essential postural tremor.

Keywords: Parkinson Disease; Essential Tremor; Wavelet; Support Vector Machine; Attention; Distraction