271 A multivariate gait data analysis technique: application to knee osteoarthritis J L Astephen* and K J Deluzio School of Biomedical Engineering, Dalhousie University, Halifax, Canada Abstract: Modern gait analysis is a powerful non-invasive tool for calculating the mechanical factors involved in pathological processes such as knee osteoarthritis (OA). Although very accurate measure- ments can be made, the clinical applicability and widespread use of gait analysis have been hindered by a lack of appropriate data analysis techniques for reducing and analysing the resulting large volumes of highly correlated gait data. This paper introduces a multidimensional gait data analysis technique that simultaneously considers multiple time-varying and discrete measures, exploiting the correlation structure between and within the measures. The multidimensional analysis technique was used to detect discriminatory mechanical features of knee OA gait patterns that involved interacting changes in several gait measures, at specific time portions of the gait cycle. The two most discriminatory features described a dynamic alignment difference and a loading response difference with knee OA. Keywords: gait analysis, data analysis, principal component analysis, discriminant analysis, knee osteoarthritis, multivariate statistics NOTATION HKA hip–knee–ankle m DP knee moment in the distal–proximal direction m LM knee moment in the lateral–medial direction a DP knee angle in the distal–proximal direction a LM knee angle in the lateral–medial direction m PA knee moment in the posterior–anterior direction a PA knee angle in the posterior–anterior direction a vector of linear discriminant function OA osteoarthritis PC principal component coefficients for features BMI body mass index PCA principal component analysis X matrix of standardized original variables c 1 stride length c 2 velocity y linear discriminant function y : classification cut-off value between two c 3 stance percentage c 4 stance time groups of discriminant scores Z matrix of principal component scores for c 5 hip–knee–angle c 6 standing knee flexion angle each subject on retained features c 7 medial joint space Subscripts c 8 body mass index DP distal–proximal coeff vector of linear discriminant function LM lateral–medial coefficients for original variables PA posterior–anterior f DP knee bone-on-bone force in the distal proximal direction f LM knee bone-on-bone force in the lateral–medial direction 1 INTRODUCTION f PA knee bone-on-bone force in the posterior–anterior direction Modern gait analysis is a powerful non-invasive tool that offers a unique means of measuring the biomechanical response to diseases of the musculoskeletal system, such The MS was received on 6 August 2003 and was accepted after revision for publication on 30 April 2004. as knee osteoarthritis (OA). The clinical applicability * Corresponding author: School of Biomedical Engineering, Dalhousie and widespread use of gait analysis, however, have been University, 5981 University Avenue, Halifax, Nova Scotia, B3H 3J5 Canada. greatly hindered by a lack of appropriate gait data analy- H05803 © IMechE 2004 Proc. Instn Mech. Engrs Vol. 218 Part H: J. Engineering in Medicine