Accelerometry data as a potential tool to help understand rare diseases

Abstract

The Wilhelmina Children’s Hospital (WKZ) provides care for a wide range of patients, including some with ultra rare diseases. Using as much data as possible is essential in treatment of such conditions. This exploratory thesis aims to investigate in what ways acceleration data can help the WKZ in understanding and treating such conditions, particularly Batten disease and Hurler syndrome. In this, we focus on analysing unlabelled acceleration data on a short time scale, which is a relatively unexplored avenue of research.

Apart from introducing a mathematical framework and information on the data, this thesis consists of two main parts. In the first part we introduce an algorithm which can locate stretches of periodic movement, and within this identifies single steps, using concepts from spectral analysis. We also introduce a tool to evaluate the performance of this algorithm, and find that tuning the parameters within the algorithm impacts the performance as expected. This makes it in itself already a broadly applicable tool.

In the second part of the thesis we analyse the recovered steps. Due to a few specific limitations we find on the available data and the current extraction algorithm, we mostly focus on cluster analysis applied to the Batten group. Doing so, we introduce a notion of connectedness, defined through the fraction of step pairs that can be seen as outliers. As this fraction is shown to be consist across several clustering techniques and parameters, and somewhat coincides with observed regularity of walking of the Batten group, this seems an interesting feature to study in more detail. In the discussion we propose a few ways of doing so, and we discuss how to potentially deal with the mentioned limitations in the analysis phase in future research.