A need for speed: Objectively identifying full-body kinematic and neuromuscular features associated with faster sprint velocities

Abstract

Sprinting is multifactorial and dependent on a variety of kinematic, kinetic, and neuromuscular features. A key objective in sprinting is covering a set amount of distance in the shortest amount of time. To achieve this, sprinters are required to coordinate their entire body to achieve a fast sprint velocity. This suggests that a whole-body kinematic and neuromuscular coordinative strategy exists which is associated with improved sprint performance. The purpose of this study was to leverage inertial measurement units (IMUs) and wireless surface electromyography (sEMG) to find coordinative strategies associated with peak over-ground sprint velocity using machine learning. We recruited 40 healthy university age sprint-based athletes from a variety of athletic backgrounds. IMU and sEMG data were used as inputs into a principal components analysis (PCA) to observe major modes of variation (i.e., PC scores). PC scores were then used as inputs into a stepwise multivariate linear regression model to derive associations of each mode of variation with peak sprint velocity. Both the kinematic (R ² = 0.795) and sEMG data (R ² = 0.586) produced significant multivariate linear regression models. The PCs that were selected as inputs into the multivariate linear regression model were reconstructed using multi-component reconstruction to produce a representation of the whole-body movement pattern and changes in the sEMG waveform associated with faster sprint velocities. The findings of this work suggest that distinct features are associated with faster sprint velocity. These include the timing of the contralateral arm and leg swing, stance leg kinematics, dynamic trunk extension at toe-off, asymmetry between the right and left swing side leg and a phase shift feature of the posterior chain musculature. These results demonstrate the utility of data-driven frameworks in identifying different coordinative features that are associated with a movement outcome. Using our framework, coaches and biomechanists can make decisions based on objective movement information, which can ultimately improve an athlete's performance.

Keywords: biomechanics; coordination; machine learning; objective movement assessment; sports performance; sprinting.

Figure 1

An example of the IMU and sEMG setup on the participant.

Figure 2

Summary of experimental workflow for both the kinematic and sEMG data set. A similar process was complete for both kinematic and sEMG data to create two different PCA frameworks.

Figure 3

(A) the distribution of peak sprint velocities for both males and females. (B) The mean sprint velocity for Male, female, and all participants.

Figure 4

(A) the velocity profile of the fastest sprinter (dark blue), slowest sprinter (light blue) and median sprinter (grey) for the entire 60 m. Peak velocity is represented by the orange dot on each velocity profile. (B) The mean and standard deviation velocity profile over the 60 m sprint and the mean and standard deviation of the position of the maximal velocity.

Figure 5

(A) scree plot for principal components derived from kinematic data, (B) scree plot for principal components derived from sEMG data.

Figure 6

Multi-component reconstruction of PCs 1, 3, 9, 11, 12 and 6 derived from kinematic data. (A) Sagittal plane view; (B) Frontal plane view and (C) Transverse plane view. The red avatar represents the 5th percentile (slow), black represents the mean and blue represents the 95th percentile (fast).

Figure 7

Multi-component reconstruction for PCs 1, 5, 21 and 22 derived from the sEMG for (A) GAS, (B) BF, (C) GMAX, (D) GMED, (E) LES, (F) LD, (G) EO, (H) VLO, (I) RF. Blue represents the 95th percentile sprinter (fast), Red represents the 5th percentile sprinter (slow), and grey represents the stance phase.