The effects of signal conditioning on the statistical analyses of gait EMG

Abstract

Ensemble averaged EMG profiles generated for leg muscles during gait have been used to clinically assess disease or injury. Several of the methods that have been reported for conditioning gait EMG signals were compared using data collected from clinically normal subjects walking on a treadmill. Specifically investigated were the effects of filtering and the quantity of data averaged upon several statistical tests that measure the variability of, or differences between, EMG profiles. Our results suggest that the variance ratio (VR) provides a reasonable test of data variability because of its modest sensitivity to both the degree of filtering and the amount of data averaged. They also suggest that of the comparison statistics: Pearson's r, the Kolmogorov-Smirnov T test and the ANOVA F ratio, the T test was the most reliable in detecting differences between given profiles for all test conditions. However, recognition of this ability of the T test must be tempered by the knowledge that while obvious EMG signal differences did exist, observable functional differences in gait did not. The relationship between statistically similar/dissimilar EMG patterns and clinically functional/dysfunctional gait patterns needs to be established. In addition, since all of the test statistics studied were affected to some degree by filtering and averaging, care should be used when comparing statistical results from separate studies unless it is known that the studies were conducted under similar conditions, including data processing. To that end, we recommend that at least 20 strides be used in the averaging process since the statistics we tested have reached or are asymptotically approaching their final values by this point.