Determination of gait patterns in children with spastic diplegic cerebral palsy using principal components Alessandra Carriero a , Amy Zavatsky b , Julie Stebbins c , Tim Theologis c , Sandra J. Shefelbine a, * a Department of Bioengineering, Imperial College London, UK b Department of Engineering Science, University of Oxford, UK c Nuffield Orthopaedic Centre, Oxford, UK 1. Introduction Three-dimensional gait analysis is a multi-factorial assessment that combines a range of measures, to provide a comprehensive description of human walking [1]. Numerous studies have exploited gait analysis to characterize normal and pathological gait patterns using a variety of gait parameters. Classifications of cerebral palsy (CP) children have been suggested, based on their ambulatory function and gait measurements. Classifications have been either qualitative, based on clinical observation, or quanti- tative, based on the cluster analysis of gait data to establish groupings. Qualitative classifications (such as jump knee, crouch knee, equinus, etc.) are most commonly used. However these are descriptive, do not use statistical techniques, and rely only on sagittal kinematic data for classification purposes [2–5]. Concerns have, therefore, been raised about their validity [5]. Other studies applied a systematic approach to classification involving different statistical clustering techniques to classify gait types based on gait analysis data [6–9]. Most of them used ‘k- means’ cluster analysis to identify gait patterns from sagittal gait analysis data over time. The k-means technique divides a set of objects into a pre-determined number of groups by maximizing the variability between different clusters and minimizing the variability within each cluster. Using this technique, O’Byrne et al. [6] identified eight groups with different gait patterns based on 237 patients with hemiplegic and diplegic CP. Kienast et al. [7] applied the k-means cluster analysis on temporal parameters and sagittal kinematic data and determined three main gait types from 14 healthy patients and 24 spastic diplegic subjects. Toro et al. [8] applied a hierarchical cluster analysis on sagittal kinematic gait data over time to 11 healthy and 56 CP children to define the optimum number of gait clusters. They detected 13 different gait patterns which were then classified as ‘crouch gait’, ‘equinus gait’ and ‘other types of gait’. O’Malley et al. [9] used ‘fuzzy’ k-means clustering on two temporal parameters: stride length and cadence (normalized respectively by leg length and age). The fuzzy clustering technique takes into account the lack of sharp boundaries between clusters: membership degrees between zero and one are used instead of crisp assignments of the data to clusters. Using this technique, O’Malley et al. classified 68 healthy and 88 spastic diplegic children in five clusters and generated an individual’s membership value for Gait & Posture 29 (2009) 71–75 ARTICLE INFO Article history: Received 4 October 2007 Received in revised form 21 May 2008 Accepted 24 June 2008 Keywords: Cerebral palsy Children Gait analysis Cluster analysis Quantitative methods ABSTRACT This study developed an objective graphical classification method of spastic diplegic cerebral palsy (CP) gait patterns based on principal component analysis (PCA). Gait analyses of 20 healthy and 20 spastic diplegic CP children were examined to define gait characteristics. PCA was used to reduce the dimensionality of 27 parameters (26 selected kinematics variables and age of the children) for the 40 subjects in order to identify the dominant variability in the data. Fuzzy C-mean cluster analysis was performed plotting the first three principal components, which accounted for 61% of the total variability. Results indicated that only the healthy children formed a distinct cluster; however it was possible to recognise gait patterns in overlapping clusters in children with spastic diplegia. This study demonstrates that it is possible to quantitatively classify gait types in CP using PCA. Graphical classification of gait types could assist in clinical evaluation of the children and serve as a validation of clinical reports as well as aid treatment planning. ß 2008 Elsevier B.V. All rights reserved. * Corresponding author at: Department of Bioengineering, Imperial College London, South Kensington Campus, Royal School of Mines Building, London, SW7 2AZ, UK. Tel.: +44 20 7594 5187. E-mail address: s.shefelbine@imperial.ac.uk (S.J. Shefelbine). Contents lists available at ScienceDirect Gait & Posture journal homepage: www.elsevier.com/locate/gaitpost 0966-6362/$ – see front matter ß 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.gaitpost.2008.06.011