17th Annual ECSS-Congress, Bruges 2012

Abstract details

Abstract-ID: 537
Session: [MO-BN03] Kinematics
Lecture room: High Live 4
Date & time: 24.06.2015 / -
Authors: Horst, F., Kramer, F., Schäfer, B., Eekhoff, A., Hegen, P., Schöllhorn, W.I.
Institution: Johannes Gutenberg University Mainz
Department: Institute of Sport Science
Country: Germany
Abstract text Introduction Biomechanical diagnoses and clinical interventions typically assume nearly constant movement patterns in their subjects. Clinical gait analysis often seeks to evaluate intervention related changes in walking by averaging the data from a number of trials and compare these average values to control subjects that did not undergo an intervention or to previous measurements in a pre-post-design. Despite the knowledge of continuous changes in living systems, movement variability without an intervention is neglected as insignificant in many approaches of movement analysis (Newell et al., 2006). The aim of this study is to look for the reliability of gait patterns from different test days by means of support vector machines. Methods Eight healthy and physically active subjects (23.5 ± 2.3 years) performed 15 gait trials at a self-selected speed on each of the eight test days within two weeks under barefoot conditions. For each trial, the continuous ground reaction forces (Kistler, 1000 Hz) and lower body joint angles (Qualisys, 250 Hz) of one gait cycle were analyzed. An eight-day-classification and one-on-one-classification of support vector machines were carried out for each subject individually. The classification rates were provided by the Liblinear Toolbox 1.4 (Fan et al., 2008) using a leave-one-out cross-validation. Results The mean classification rates for the eight-day-classification are sagittal (71.4 ± 10.4%), frontal (90.7 ± 8.4%), coronal (92.1 ± 8.2%) and combined (95.9 ± 5.8%) joint angles. The mean classification rates for the eight-day-classification are fore-aft (49.6 ± 9.2%), medial-lateral (49.5 ± 9.9%), vertical (48.4 ± 8.7%) and combined (60.1 ± 9.2%) ground reaction forces. The mean classification rates for the one-on-one-classification are sagittal (88.7 ± 5.7%), frontal (95.9 ± 2.2%), coronal (96.9 ± 2.1%) and combined (98.1 ± 1.1%) joint angles. The mean classification rates for the one-on-one-classification are fore-aft (82.5 ± 7.7%), medial-lateral (80.9 ± 6.1%), vertical (83.2 ± 8.2%) and combined (86.1 ± 6.7%) ground reaction forces. Discussion The eight-day-classification rates of 95.9% and 60.1% clearly differ from a random classification of 12.5% and show natural differences between the gait patterns of different days. Hence, changes in gait patterns appear naturally without a specific intervention between the test days. Additionally, the one-on-one-classification points out a general problem of studies with pre-post-design. References Fan RE, Chang KW, Hsieh CJ, Wang XR, Lin CJ (2008).Journal of Machine Learning Research, 9, 1871-4. Newell KM, Deutsch KM, Sosnoff JJ & Mayer-Kress G (2006). Movement system variability, 3-23. Human Kinetics, Champaign (Ill). Contact horst@uni-mainz.de
Topic: Biomechanics
Keyword I: gait patterns
Keyword II: reliability
Keyword III: support vector machines