Activity Recognition to Improve Motor Performance in Parkinson’s Disease

Through sensors worn on the body or embedded into objects of daily use we can infer the activities performed by a subject. Extracting the characteristics of the data collected by these sensors, i.e. how these activities were performed, would be beneficial to a variety of applications, such as rehabilitation, pain therapy, sports and professional training in tool usage, e.g. for mechanics among others. Information about the development of motor performances and whether there is an improvement or decline over time can be very useful, particularly in medicine and specifically in degenerative conditions such as Parkinson’s Disease, where an assessment of a decline in motor ability is a common diagnostic tool. So far, however, relatively little work has been invested into further, detailed analysis of daily activities.

In this project we aimed to develop a method that could be used to assess the efficiency of motion, one of the properties of motor skill. This method could be applied for people with degenerative conditions that have a significant impact on motor abilities, such as Parkinson’s Disease and Dementia. Our method for measuring motor efficiency was based on the energy distribution in Principal Component Analysis (PCA) and we used it to infer a single, normalised metric that was intimately linked to signal complexity and allowed comparison of (subject-specific) time-series. We evaluated the approach on artificially distorted signals and applied it to a simple kitchen task to show its applicability to real-life data streams.

Date: Oct 2010 – Aug 2011

Funding: EPSRC: Engineering and Physical Sciences Research Council (KTA) £51,704

Researchers: Patrick Olivier (PI). Richard Walker – Institue of Health & Society, Nick Miller – School of Education, Communication, and Language Sciences, Lynn Rochester – Institute of Aeging & Health (CIs). Roisin McNaney, Karim LadhaThomas Ploetz, Nils Hammerla, Dan Jackson.