Classification of Ataxic Gait

Abstract

Gait disorders accompany a number of neurological and musculoskeletal disorders that significantly reduce the quality of life. Motion sensors enable high-quality modelling of gait stereotypes. However, they produce large volumes of data, the evaluation of which is a challenge. In this publication, we compare different data reduction methods and classification of reduced data for use in clinical practice. The best accuracy achieved between a group of healthy individuals and patients with ataxic gait extracted from the records of 43 participants (23 ataxic, 20 healthy), forming 418 segments of straight gait pattern, is 98% by random forest classifier preprocessed by t-distributed stochastic neighbour embedding.

Keywords: SARA; ataxia; classification; gait; machine learning.

Figure 1

Flow diagram of data processing.

Figure 2

Principal component analysis is a tool that ranks the features by the eigenvalues, which corresponds to the desired ability of discrimination between classes (A). Of all the techniques compared, it has the worst ability to categorise individual participants (B).

Figure 3

The uniform manifold approximation and projection model can be used for predicting a new sample to group (A) or similarity to the other study participant (B).

Figure 4

Classifier decision surface plotted for selected classifiers in dimension of two T-distributed stochastic neighbour embedding components. The achieved accuracy of classification for each specific case is shown in the yellow box in the right bottom corner.

Figure 5

T-distributed stochastic neighbour embedding is also able to well separate subjects according to the severity of the clinical disability measured by the Scale for the Assessment and Rating of Ataxia score (the higher score the more severe ataxia).