The influence of walking speed and effects of signal processing methods on the level of human gait regularity during treadmill walking

Abstract

Abstract Background In recent years the use of sample entropy (SampEn) to evaluate the complexity of the locomotor system in human gait data has gained in popularity. However, it has been suggested that SampEn is sensitive to various input parameters and signal preprocessing methods. This study quantified the effects of different temporal and spatial normalization approaches and various lengths of the template vector (m) on SampEn calculations. The discriminatory ability of SampEn was studied by comparing two walking conditions. Methods Twenty-three participants (seven males, 55.7 ± 8.5 years, 165.7 ± 7.9 cm, 80.5 ± 16.7 kg) walked on a treadmill with preferred (Vpref) and maximum (Vmax) speed. Data were segmented and resampled (SEGM), resampled and spatially normalized (NORM), resampled and detrended (ZERO). Results For vertical ground reaction force (vGRF) and center of pressure in anterio-posterior direction (COPap), in both walking conditions, SampEn was generally sensitive to the vector length and not to the data processing, except for COPap in ZERO, m = 2, 4. For the COPml SampEn behaved oppositely, it was sensitive to preprocessing method and not to the m length. The regularity of COPap and vGRF in all processed signals increased in Vmax condition. For the COPml only two signals, WHOLE and ZERO, revealed increased complexity caused by more demanding walking conditions. Conclusions SampEn was able to discriminate between different walking conditions in all analyzed variables, but not in all signals. Depending on evaluated variable, SampEn was susceptible in different way for the m level and processing method. Hence, these should be checked and selected for each variable independently. For future studies evaluating influence of walking velocity on COP and vGRF regularity during treadmill walking it is advised to use raw time series. Furthermore, to maintain template vector which represents biological relevance it is advised to detect highest frequencies present in analyzed signals and evaluate minimal time interval which can reflect change caused by response of a neuromuscular system. During evaluating treadmill walking measured with 100 Hz sampling frequency it is recommended to adopt m from 6 to 10, when average stride time is up to about 1 s.