A method for the classification of corrective activity in context dependent postural controlling tasks

Abstract

Difficulties in maintaining postural stability are not common among young healthy people. However, with increasing age problems start to emerge. Deficits in the postural control system may also originate from a working environment where noise and solvents are present. Some diseases, for instance Menière's disease, can cause instability in walking and standing. Regardless of the origin of the problem in the balance system, it has to be detected in a meaningful, easily interpretable way. When detected, a suitable rehabilitation method can be proposed. In this paper we present a method which extracts a scalar feature from a stabilogram signal, which well describes the differences between young and elderly people's swaying processes. When our feature is applied to the K nearest neighbour algorithm the correct recognition accuracy is over 70% in cases where the purpose is to predict whether an unknown test feature is measured from a young or an elderly subject.

Introduction

Usually humans do not have to concentrate on their walking, standing and other balance related tasks. However, due to vertigo, certain diseases and aging, these tasks may become difficult or even impossible. If this is the case, the next step is to explore what the possible reasons could be. In addition, proper rehabilitation methods should be studied. In order to find the underlying reasons of impaired ability to maintain balance we need ample medical expertise and reliable methods to measure the human swaying process. Of course, many of these reasons are natural and they cannot be affected. If some of the reasons could be resolved, the quality of life of many people could be improved. For instance, the number of hip fractures among the oldest of the old could be reduced.

The investigation of the human postural control system is a very complex task and can be approached from many different viewpoints. The interest is usually physiological [1], [2], [3], medical [4], [5], [6] or psychological [7], [8], [9]. In all these situations we must have a evidence-based way to measure the human ability to maintain an upright stance. Obviously, the most common method to measure human swaying is the use of the force platform [10], [11], [12]. The use of video [13], [14], [15] and inertial based systems [16], [17] has also been proposed.

Measured data are usually “raw” from the application point of view and should be somehow pre-processed. The main purpose of the pre-processing stage is to identify such features from the original data which best characterize the problem under study. In addition, the dimension of these features should be small as to gain the best achievable computational efficiency.

Many features have been evinced as explanations for the nature of human swaying, among them diffusion analysis, fractal analysis and time series analysis [18], [19], [20], [21], [22]. However, researchers do not necessarily agree on the usefulness of these. Furthermore, some of these features are usually suitable for machine learning, but for humans their interpretation is difficult.

In this paper we present a pre-processing method which extracts information from stabilogram signals. This method provides a feature which is efficient for machine learning purposes and its interpretation for humans is intuitive. In addition, we present classification results of the K-nearest neighbour algorithm which uses our feature values.