Knee adduction and flexion moments do not reflect knee loading during stance phase of gait

Abstract

s / Gait & Posture 42S (2015) S1–S90 S59 decreased IGFBP-3 concentrations compared to control in bothpreand post-training. http://dx.doi.org/10.1016/j.gaitpost.2015.03.103 Pattern recognition methods in clinical gait analysis – What do we gain? M. Bruderer-Hofstetter1,∗, F. Rast2, C. Bauer2, E. Graf2, A. Meichtry1 1 Institute of Physiotherapy, University of Applied Sciences Zurich, Winterthur, Switzerland 2 Laboratory for Motion Analysis, University of Applied Sciences Zurich, Winterthur, Switzerland Introduction and aim: Pattern recognition methods have been widely used in clinical gait analysis [1]. Although a combination of principal component analysis (PCA) and support vector machines (SVM) demonstrated a high sensitivity to visualize subtle differences inmovement patterns [2,3] the benefit of pattern recognition methods in clinical application remains uncertain. Therefore, the aim of our study was to compare the results of a discrete parameter analysis with a pattern recognition approach in order to detect differences in kinematics and kinetics between two different shod conditions. Patients/materials and methods: Five walking trials for each of the two conditions of 20 healthy subjects were captured using a 7-camera-motion-analysis-system (Vicon, 200Hz) and two forceplates (AMTI, 1000Hz). Discrete parameter analysis calculated a priori chosen kinematic and kinetic parameters. Pattern recognition approach (PCA) extracted gait features (PC) from an input matrix containing kinematic and kineticwaveforms. Subsequently, a SVM classifier with linear kernel function was applied to determine the optimal PC-subspace. Within the remaining PCs and all parameters paired sample t-testswere applied and Cohens’ d effect sizes were calculated. Results:Discrete parameter analysis revealed significant differences (p<0.01) for the second peak ankle flexion angle (d=0.60), the peak accelerating ground reaction force (GRF) (d=0.86) and the second peak vertical GRF (d=−0.47). Pattern recognition approach detected four condition dependent PCs. Significant differences (p<0.01) were found in PC2 and PC3 (effect sizes d=−1.21 and d=2.75, respectively). PC2 corresponded to angle differences at mid-stance at hip, knee and ankle joints and to anterior-posterior and vertical GRF differences at terminal-stance. PC3 represented knee angle difference at mid-stance, anterior-posterior GRF differenceat terminal-stanceandangledifferencesathipandankle joints and vertical GRF difference at pre-swing. Discussion and conclusions: With the pattern recognition approach a deeper insight into the data was achieved. Differences between the two shod conditions were detected in other sections of the gait cycle than a priori supposed. Although the pattern recognition approach detected condition dependent differences independently of prior knowledge the benefit for clinical gait analysis application hinges on the ability to interpret the extracted gait features.