Surface EMG signal normalisation and filtering improves sensitivity of equine gait analysis

Abstract

Low-frequency noise attenuation and normalisation are fundamental signal processing (SP) methods for surface electromyography (sEMG), but are absent, or not consistently applied, in equine biomechanics. The purpose of this study was to examine the effect of different band-pass filtering and normalisation conventions on sensitivity for identifying differences in sEMG amplitude-related measures, calculated from leading (LdH) and trailing hindlimb (TrH) during canter, where between-limb differences in vertical loading are known. sEMG and 3D-kinematic data were collected from the right Biceps Femoris in 10 horses during both canter leads. Peak hip and stifle joint angle and angular velocity were calculated during stance to verify between-limb biomechanical differences. Four SP methods, with and without normalisation and high-pass filtering, were applied to raw sEMG data. Methods 1 (M1) to 4 (M4) included DC-offset removal and full-wave rectification. Method 2 (M2) included additional normalisation relative to maximum sEMG across all strides. Method 3 (M3) included additional high-pass filtering (Butterworth 4th order, 40Hz cut-off), for artefact attenuation. M4 included the addition of high-pass filtering and normalisation. Integrated EMG (iEMG) and average rectified value (ARV) were calculated using processed sEMG data from M1 – M4, with stride duration as the temporal domain. sEMG parameters, within M1 – M4, and kinematic parameters were grouped by LdH and TrH and compared using repeated measures ANOVA. Significant between-limb differences for hip and stifle joint kinematics were found, indicating functional differences in hindlimb movement. M2 and M4, revealed significantly greater iEMG and ARV for LdH than TrH (p<0.01), with M4 producing the lowest p values and largest effects sizes. Significant between-limb differences in sEMG parameters were not observed with M1 and M3. The results indicate that equine sEMG SP should include normalisation and high-pass filtering to improve sensitivity for identifying differences in muscle function associated with biomechanical changes during equine gait.