Modular Control of Treadmill vs Overground Running

Abstract

Motorized treadmills have been widely used in locomotion studies, although a debate remains concerning the extrapolation of results obtained from treadmill experiments to overground locomotion. Slight differences between treadmill (TRD) and overground running (OVG) kinematics and muscle activity have previously been reported. However, little is known about differences in the modular control of muscle activation in these two conditions. Therefore, we aimed at investigating differences between motor modules extracted from TRD and OVG by factorization of multi-muscle electromyographic (EMG) signals. Twelve healthy men ran on a treadmill and overground at their preferred speed while we recorded tibial acceleration and surface EMG from 11 ipsilateral lower limb muscles. We extracted motor modules representing relative weightings of synergistic muscle activations by non-negative matrix factorization from 20 consecutive gait cycles. Four motor modules were sufficient to accurately reconstruct the EMG signals in both TRD and OVG (average reconstruction quality = 92±3%). Furthermore, a good reconstruction quality (80±7%) was obtained also when muscle weightings of one condition (either OVG or TRD) were used to reconstruct the EMG data from the other condition. The peak amplitudes of activation signals showed a similar timing (pattern) across conditions. The magnitude of peak activation for the module related to initial contact was significantly greater for OVG, whereas peak activation for modules related to leg swing and preparation to landing were greater for TRD. We conclude that TRD and OVG share similar muscle weightings throughout motion. In addition, modular control for TRD and OVG is achieved with minimal temporal adjustments, which were dependent on the phase of the running cycle.

Fig 1. Experimental design.

Illustration of experimental design involving treadmill (blue) and overground running (red) from a representative subject. Tibia vertical acceleration used to identify gait cycles, and surface EMG are plotted in (A) (raw data for acceleration, envelope for sEMG). The gait cycles were defined from one left foot strike to the following left foot strike. The motor modules extracted from the concatenated surface EMG are represented as “muscle weightings” (panel B) and “activation signals “(panel C), in this case 5 consecutive gait cycles are reported. It is worth noting that by concatenating several gait cycles we accounted the inter-cycle variability in the analysis.

Fig 2. Schematic representation of EMG data processing.

EMG data from treadmill (TRD) and overground running (OVG) were band-pass filtered, low-pass filtered and segmented into (20) running cycles. These segmented data were processed for the extraction of peak EMG and integrated EMG. In addition, the filtered and segmented data were processed using non-negative matrix factorization (NMF) in order to extract muscle weightings and respective activation signals for TRD and OVG separately. We reconstructed the original EMG from OVG (OVG EMG) by mixing muscle weightings and activation signals from TRD condition, generating a reconstructed OVG-REC_WA. A second mixed model was the reconstruction of the original EMG from OVG by mixing TRD muscle weightings and OVG activation signals, generating a reconstructed OVG-REC_W. The third mixed model was the reconstruction of the original EMG from OVG by mixing OVG muscle weightings and TRD activation signals, generating a reconstructed OVG-REC_A. This same procedures were applied for the reconstruction of the original EMG from TRD.

Fig 3. EMG Comparison of treadmill vs overground.

Percentage of change between treadmill (TRD) and overground (OVG) running for the peak EMG (A) and integrated EMG (B) throughout a gait cycle. Data from OVG running were normalized by the values from TRD running. * denotes significant differences between treadmill and overground running (p<0.05).

Fig 4. Motor modules—Running.

Reconstruction quality of surface EMG signals (variation accounted for—VAF) by means of different number of motor modules (A) for each subject separately. The horizontal solid line in each panel corresponds to VAF at 0.9, and the vertical solid line correspond to the extraction of four motor modules. Mean±SD VAF for each condition and EMG processing method is displayed in the panels. We did not observe differences between locomotion conditions (treadmill vs overground running). In B, motor modules extracted from 20 consecutive gait cycles from treadmill running (blue) and overground running (red) are displayed for all subjects (individual bars of the muscle weightings and lines of the activation signals).

Fig 5. Motor modules—similarities.

A: inter-subject similarities (Mean±SD) across the four identified muscle weightings and activation signals during treadmill running (grey) and overground running (black). B: intra-subject similarities between treadmill and overground running. * denotes significant difference in relation to overground running.

Fig 6. Modular organization—treadmill vs overground.

A: Mean (black line) and SD (gray area) absolute subtraction of treadmill (TRD) from overground (OVG) activation signals. B: mean (SD) timing of the peak amplitude of activation signals from TRD and OVG running for each motor module (from M1 to M4). C: mean (SD) peak amplitude of activation signals for each motor module (from M1 to M4). D: reconstruction quality of multi-muscle EMG datasets from different combinations of muscle weightings and activation signals (please refer to Methods for explanation). † denotes significant difference in relation to TRD running (p<0.05). * denotes significant difference in relation to Regular and REC_W only (p<0.001).