Emerging portable technology provides highly predictive models of sprint performance in elite Australian track athletes: a foundation for measuring sprint performance in the field
Title
Authors
A
TimothyA.Sayer1,2✉Emailtimothy@melbournecbdsportsmed.com.au
XavierEvans1
NicholasCross1,3
Dr.
TimSayer41Melbourne CBD Physiotherapy and Sports Medicine ClinicMelbourneVictoriaAustralia
2Department of PhysiotherapyThe University of MelbourneMelbourneVictoriaAustralia
3Department PhysiotherapyLa Trobe UniversityMelbourneVictoriaAustralia
4Melbourne CBD Physiotherapy and Sport MedicineGround Floor, 179 Queen St3000MelbourneAustralia, Australia
Timothy A. Sayer1,2 Xavier Evans1 & Nicholas Cross1,3
1. Melbourne CBD Physiotherapy and Sports Medicine Clinic, Melbourne, Victoria, Australia
2. Department of Physiotherapy, The University of Melbourne, Melbourne, Victoria, Australia
3. Department Physiotherapy, La Trobe University, Melbourne, Victoria, Australia
Corresponding Author
Dr. Tim Sayer
Melbourne CBD Physiotherapy and Sport Medicine Australia
Ground Floor, 179 Queen St
Melbourne 3000, Australia
Email: timothy@melbournecbdsportsmed.com.au
Abstract
Background
The purpose of this study was to identify the key predictors of sprint performance in elite Australian track and field athletes using portable field-based technology.
Methods
Twenty-eight elite Australian track and field sprinters (16 male, 12 female, 69.21 ± 8.45 kg, 1.74 ± 6.88 metres) completed 2 x 60 metre sprints recorded via a MuscleLab LaserSpeed, SportScientia Techlayer and VueMotion kinogram.
Results
For 60 metre sprint time, the regression model was highly significant, F(5, 50) = 136.20, p < 0.001, explaining 93.2% of the variance (R² = 0.932), with maximal velocity (MaxV) as the strongest predictor (β = − 1.118, p < 0.001) and MaxStepFreq also significant (β = 0.210, p = 0.017). Prediction of MaxV across 60 metres was also strong (R² = 0.885, p < 0.001), with MaxStepFreq (β = 0.600) and min_accel_magnitude (β = 0.402) as key predictors. For the 20 metre sprint time, the model accounted for 74.7% of variance (R² = 0.747, p < 0.001), with MaxV_20m the only significant predictor (β = − 0.869, p < 0.001) and a subsequent model for MaxV_20m was also significant explaining 59.0% of the variance in MaxV_20m (R² = .590, Adjusted R² = .548) with the percent_maxV60 as the only statistically significant contributor (β = −0.237, p = .043).
Conclusion
These findings suggest that force-velocity metrics derived from portable instrumented insoles and infrared radar are critical for measuring sprint performance. This provides sport scientists and health clinicians a strong foundation for tracking performance changes in elite sprinting.
Keywords:
sprint performance
sprint biomechanics
maximum velocity
portable technology
Competing Interests
The authors declare that they have no competing interests" in this section.
A
Funding
No funding was received from government grants or organizations. All research was funded by the authors business privately or given in kind.
Ethics Declaration
All participants in this study signed a consent form as approved by the University of Melbourne human ethics research committee.
A
A
A
Introduction
A
Sprint performance is critical for success in both track and field sprint events and various field-based sports. Sprinting is a highly complex motor task influenced by a range of neuromuscular, biomechanical, and anatomical factors 1,2. Understanding which factors are critical to sprint performance has long been desired by coaches, sports scientists and health professionals, as key predictors of sprint performance could lay the foundation for tracking athletic progress of targeted interventions. Moreover, the identification of key sprint predictors may provide insight for methods to reduce soft tissue injuries such as hamstring tears, given that many of these injuries occur during maximal sprint efforts 3.Regression-based analyses across multiple studies have shown that horizontal ground reaction force and net horizontal impulse are among the strongest predictors of sprint performance 1. Morin et al.1 recruited 12 sub-elite sprinters and examined maximal sprinting on an instrumented treadmill compared to 100 metre sprint time and found that ratio of force and net horizontal ground reaction force were strongly correlated (R² = 0.64) to maximal velocity. Further evidence from Cross et al.4 systematic review highlights that peak horizontal power has also shown to strongly correlate with sprint time, and kinematic variables such as step length, step frequency, and contact time also demonstrate predictive value, although with slightly lower R² values (0.54-058). These studies highlight that sprint performance is strongly correlated with biomechanical variables associated with foot-ground interaction and the application of force quickly to achieve maximal speed.
Despite these findings, consensus on the optimal set of predictors remains elusive, partly due to methodological variability across studies due to sub-elite populations, small sample sizes, and a lack of field-based data collection methods outside traditional 3-dimensional motion analysis laboratories. Recent technological advancements, such as wireless insoles with integrated inertial measurement units (IMU) and force sensors, high-frequency radar and lower body kinograms via mobile video footage now enable real-time, in-field assessment of sprint biomechanics. Hence, the primary aim of this study was to characterize and develop key predictors of sprint performance and acceleration using portable technology among elite Australian track sprinters.
Materials and Methods
Study Design
A
This was a cross-sectional observational study designed to identify biomechanical predictors of maximal sprint performance in elite track and field athletes. A field-based assessment was conducted using wearable sensor technology, laser timing, and motion analysis tools across standardized sprint trials. Twenty-eight elite Australian track national-level competitors and Olympic qualifier sprinters (mixed sample: 16 male, 12 female, 69.21 ± 8.45 kg, 1.74 ± 6.88 m) participated in this study. Inclusion criteria were Australian national level competition status competing in 100, 200 or 400 metre sprints and no lower-limb injury within 6 months prior to testing. A
A
All participants provided informed consent in accordance with ethics approval of the University of Melbourne by the human ethics research committee (ethics ID 30942).A
Sprint ProtocolAthletes completed their regular warm-up routines as per their individual training practices. Each athlete then performed two maximal 60-metre sprints on a Mondo athletics track, starting from a standardized three-point position. Instructions were given to the athlete to begin on the start line and from a three-point stating position to begin in their own time. This was chosen to minimize the influence of reaction time on sprint performance. Adequate rest (> 10 minutes) was provided between sprint efforts to minimize fatigue influencing their maximal sprint efforts.
Timing and maximal velocity
Sprint time and maximal velocity across 60 metres and 20 metres were measured using a MuscleLab LaserSpeed system (Ergotest, Norway) with established validity and reliability measuring sprint performance 5. The radar was positioned 10 metres behind the starting line and sampling internally at 2.56 kHz with a 20 Hz output. The radar was set-up with the use of MuscleLab software and set to auto-detect mode for each sprint whereby recording would begin once the athlete’s centre of mass was moving away from the radar at > 0.01 m/s. All athletes wore a reflective vest to ensure accurate data capture. The device recorded time in seconds (s) to 60 metres and segmented data into intervals of 10 metres. For the 60 metre data the maximal velocity across 60m (MaxV, m/s) and maximal step length x step frequency (MaxStepFreq, m•Hz) was extracted. For the 20 metre acceleration, the 20 metre maximal velocity (MaxV_20m), percentage of maximal velocity at 20 metres relative to 60 m metres (percent_maxV60) was extracted for each sprint from the MuscleLab software using their custom algorithms. Data was extracted by Microsoft Excel CSV files for each athlete.
Spatiotemporal and force-time variables
A wireless Techlayer insole (SportScientia PTE. LTD incorporated, Singapore), was inserted into each athlete’s sprint spike before the sprint. This thin (1–5 mm) and lightweight (45 g) insole integrates a 6-axis Inertial Measurement Unit (IMU) composing a 16G accelerometer (500 Hz) a 200G high-acceleration sensor (500 Hz) and a gyroscope (2000 dps at 500 Hz). In addition, the Techlayer insole has nine force sensors throughout the forefoot and heel of the insole (Fig. 1). The Techlayer houses a wireless battery which was charged before each session using a SportScientia charging box that mounts to the battery in the middle of the insole where the IMU is located. Once charged, the Techlayer insole was paired and synced with the SportScientia software on a laptop at the track and placed inside the spike of each athlete. The weight of the athlete walking automatically cued the recording of data from each insole (left and right side) through the detection of movement and pressure on the force sensors and IMU. Following each sprint, the Techlayer was removed from the spike and mounted to the charging box and connected to the internet whereby data was uploaded to the SportScientia cloud data hub using their patented software. Each sprint was identified using a custom SportScientia algorithm that identified peak force (N), peak force x contact time (peakforce_CT, N•ms) contact time (CT, ms), impulse (kN/s), vertical loading rate (VLR, N/s), peak speed (m/s), time between ipsilateral initial contact (IC, ms), minimum acceleration magnitude (min_accel_magnitude, G), time to peak force (ms ) and swing speed (m/s) throughout the sprint. Every step across the 60 metre sprint was collated and averaged for each variable across 0–60 metres and 0–20 metres for analysis and exported to a Microsoft Excel CSV file.
Fig. 1
Schematic illustration of SportScientia Techlayer. A thin (1–5 mm) and lightweight (45 g) insole integrates a 6-axis Inertial Measurement Unit (IMU) composing a 16G accelerometer (500 Hz) a 200G high-acceleration sensor (500 Hz) and a gyroscope (2000 dps at 500 Hz).
Kinematics
Kinematics were collected using VueMotion Labs (Sydney, Australia) kinograms over the final 20 metres of each sprint using an iPhone 15 Plus (Apple Incorporated) mounted to a tripod capturing the sagittal plane running of each athlete at 60Hz. The video footage of each sprint from 40–60 metres was downloaded from the iPhone library and uploaded to VueMotion software and analysed using its proprietary algorithm whereby a calibration method is used to accurately map out ground-based measures. The knee angle at full support in the stance phase (knee_angle_full_support, °), thigh angular velocity (TAV, rad/s) and thigh split angle at initial contact (IC, °) were extracted and reported as means for the 60 metre sprint within the 40–60 metre capture zone. VueMotion data was not utilised for the 0–20 metre analysis as there was no data capture in this phase of the sprint.
Data Analysis
All variables were extracted for statistical modelling with descriptive statistics (mean ± SD) calculated and reported for both 60 and 20 metre sprints. A total of 21 variables (Table 1) were collected from two repeated sprints in each subject. Due to the sample size, we conducted a preliminary step before multiple linear regression modelling for predictors of sprint performance. This first step was to utilise Python code script via Google Collaboratory to fit a random forest regressor, to predict the dependent variable (time) in model 1 with all 18 variables as independent predictors. A random forest regressor predictive model was chosen because it can adequately handle multicollinearity between variables within the dataset and will provide sufficient insight into variable importance with proper post-hoc analysis. This machine learning approach split the dataset into 80% for training the model, and the remaining 20% for prediction comparison to provide model accuracy. A SHapley Additive exPlanations (SHAP explainer) was created for post-hoc analysis of the model, with Beeswarm plots generated to highlight the most influential variables. This final step produced a rank order of all 18 independent predictors for time as the dependent variable. This process was performed again for model 2 with maximal velocity as the dependent variable and again for model 3 and model 4 for the 20 metre analysis. Subsequently, the top five rank order of independent predictors for each model were selected and put into a multiple linear regression model via SPSS v27.0 (IBM Corp, Armonk, NY, USA). the enter method was utilised following insight provided by the random forest model of the most important variables for prediction. Significance was set at p < 0.05 and model fit was evaluated using R², adjusted R², and standard error of the estimate (SEE).
Technology | Variables | 0-60m (mean ± SD) | 0-20m (mean ± SD) |
|---|---|---|---|
MuscleLab Radar | Time (s) | 7.52 ± 0.42 | 3.27 ± 0.17 |
MaxV (m/s) | 9.67 ± 0.66 | - | |
MaxV_20m (m/s) | - | 8.84 ± 0.46 | |
MaxStepFreq (m•Hz) | 9.37 ± 0.72 | - | |
Percent_MaxV60 | - | 91.46 ± 2.14 | |
SportScientia Techlayer | Peak force (N) | 11603.37 ± 1727.31 | 9674.79 ± 1307.14 |
Peak force (N/kg) | 168.60 ± 24.25 | 140.61 ± 18.30 | |
Peakforce_CT (N•ms) | 102.79 ± 19.61 | 73.87 ± 12.64 | |
min_accel_magnitude (G) | 37.10 ± 10.16 | 24.24 ± 6.14 | |
Time to peak force (ms) | 67.87 ± 7.68 | 81.80 ± 9.34 | |
CT (ms) | 113.86 ± 6.69 | 131.97 ± 8.44 | |
Impulse (kN.S) | 0.84 ± 0.09 | 0.79 ± 0.09 | |
VLR (N/s) | 195396.62 ± 48156 | 139512.62 ± 29280.70 | |
Peak speed (m/s) | 9.30 ± 0.67 | 9.30 ± 0.66 | |
Peak swing speed (m/s) | 15.16 ± 1.07 | 14.57 ± 0.96 | |
Time between ipsilateral limb IC to IC (ms) | 461.24 ± 16.41 | 457.83 ± 18.40 | |
knee_angle_full_support (º) | 149.92 ± 5.99 | - | |
VueMotion Kinograms | TAV (rad/s) | 9.64 ± 0.58 | - |
Thigh split angle at IC (º) | 7.66 ± 15.32 | - | |
Positive running score | 22.25 ± 4.37 | - | |
IC distance to COM (cm) | 14.68 ± 3.74 | - | |
| CT = contact time, IC = initial contact, MaxStepFreq = maximum step length * maximum step frequency, min_accel_magnitude = minimum acceleration magnitude, Peakforce_CT = peak force *contact time, SD = standard deviation, TAV = thigh angular velocity, VLR = vertical loading rate | |||
| - denotes that no data was calculated for this variable across the distance | |||
Findings/Results
60 metre sprint analysis
A
The first multiple linear regression model for prediction of time was statistically significant, F(5, 50) = 136.20, p < .001, accounting for 93.2% of the variance in sprint time (R²= .932, adjusted R²= .925, Fig. 2). The SEE was 0.118, indicating strong model fit. Among the predictors, MaxV emerged as the most influential variable (β= −1.118, p < .001), with higher velocities strongly associated with shorter sprint times (Fig. 2).The MaxStepFreq was also a significant positive predictor (β = 0.210, p = .017). In contrast, knee_angle_full_support (β = 0.054, p = .216), min_accel_magnitude (β = 0.001, p = .984), and CT (β = 0.081, p = .133) did not significantly contribute to the model.The Pearsons correlation analyses demonstrated strong negative relationships between sprint time and both MaxV (r = − 0.958, p < .001) and MaxStepFreq (r = − 0.817, p < 0.001) highlighting the very strong association between time across 60 m and MaxV and MaxStepFreq (Fig. 2). Furthermore, sprint time also had strong negative correlation with min_accel_magnitude (r= − 0.803, p < .001), indicating that greater minimum acceleration values have association with faster sprint times. Likewise, a moderate positive correlation was observed between sprint time and CT (r = 0.650, p < .001), indicating that shorter ground contact time are associated with quicker sprint times.
A
The second multiple linear regression model which analysed predictors of MaxV (Fig. 3), as it was the strongest predictor of time across 60 metres, was statistically significant, F(5, 50) = 76.74, p < 0.001, explaining 88.5% of the variance in MaxV (R²= 0.885, adjusted R²= 0.873). The SEE was 0.236, indicating a strong fit between the predicted and observed values. Among the predictors, MaxStepFreq emerged as the strongest contributor (β = 0.600, p < 0.001), followed by min_accel_magnitude (β = 0.402, p < 0.001). However, the remaining predictors peak force(β = 0.381, p = 0.128), peakforce_CT (β= −0.416, p = 0.161) and time to peak force (β= −0.106, p = 0.198) did not significantly contribute to the model.Pearson correlations revealed significant associations between MaxV and all independent variables. MaxV was most strongly correlated with MaxStepFreq (r = 0.894, p < 0.001), min_accel_magnitude (r = .824, p < .001), peakforce_CT (r = 0.753, p < 0.001), and peak force (r = 0.712, p < 0.001). Notably, time to peak force was negatively correlated with MaxV (r = − 0.556, p < 0.001), suggesting that shorter time to peak force is associated with higher sprint velocity.
20 metre acceleration analysis
Results for the multiple linear regression analysis of 20 metre sprint time as the dependent variable was statistically significant, F(5, 50) = 29.57, p < 0.001, accounting for 74.7% of the variance in sprint time (R² = 0.747, Adjusted R² = 0.722), with an SEE of 0.091. Among the predictors, MaxV_20m was the strongest and only statistically significant contributor (β = − 0.869, p < 0.001). Peak force (β = 0.285, p = 0.123), impulse (β = − 0.268, p = 0.083), CT (β = 0.201, p = 0.137), and time to peak force (β = − 0.096, p = 0.476) did not reach statistical significance.
A
Pearson correlation revealed a strong negative association between 20 metre sprint time and MaxV_20m (r= − 0.851, p < 0.001, Fig. 4), indicating that higher maximal sprint velocity was closely linked to faster sprint performance. Peak force (r= − 0.598, p < 0.001) and impulse (r= − 0.526, p < .001) also demonstrated moderate negative correlations with sprint time.A
A subsequent model was then analysed for MaxV_20m which was also statistically significant (Fig. 5), F(5, 50) = 14.36, p < 0.001, explaining 59.0% of the variance in MaxV_20m (R² = .590, Adjusted R² = .548). The SEE was 0.314, indicating a reasonably good fit. Among the predictors, percent_maxV60 was the only statistically significant contributor (β = −0.237, p = .043). The minimum acceleration magnitude, peakforce_CT, swing speed, and VLR showed moderate to high standardized beta coefficients (β = 0.110–0.265), yet none reached statistical significance (p > 0.05).Pearson correlation analysis revealed several significant associations with MaxV_20m. The strongest positive correlations were observed with VLR (r = 0.712, p < 0.001), peakforce_CT (r = 0.684, p < 0.001), and minimum acceleration magnitude (r = 0.561, p < 0.001). Conversely, percent_maxV60 was moderately negatively correlated with MaxV_20m (r= − 0.546, p < 0.001),
Discussion
This study is the first to identify key predictors of time in elite Australian track sprinters across 60 and 20 metres using portable technology collected in the field. We found that MaxV was the most influential predictor of 60 metre sprint performance, explaining over 93% of the variance. Moreover, the correlation between maximal velocity and time was also significantly high at 95.8% (r = − 0.958). This is significant as it is the first study to demonstrate such a high predictive value utilising emerging portable technology in contrast to traditional lab motion analysis systems. Our research extends previous findings by Healy et al.6 whereby maximal velocity was correlated with 100 metre sprint times (r= -0.97) across international events reconstructed from video footage to derive a time x velocity curve. Moreover, Tam and Yao7 utilized machine learning models to analyse elite-level 100 metre sprint data and reported a strong negative correlation between MaxV and overall sprint time, confirming that athletes who attain higher speed achieve quicker times. Importantly, these findings reinforce the concept that achieving a higher maximal velocity is fundamental to sprint performance, and may provide new evidence for coaches, sport scientists and health professionals to measure in the field for tracking sprint performance and success of training interventions 8.
The second key predictor in our study was MaxStepFreq, which was a combined variable of maximum step length and step frequency and consequently was shown to be significant to both sprint time and maximal velocity. This is similar to the findings of Morin et al 1, who showed that step frequency through a higher cadence was achieved through shorter ground contact times among 12 active males. These results are supported by earlier observations by Weyand et al. 9, who demonstrated that higher velocities on an instrumented treadmill are achieved by applying greater force in less time, thereby enabling a quicker step rate (frequency). Our study extends these studies as we measured the maximal step length and frequency in the field from a radar and completed this in a larger cohort of elite track sprinters (n = 28). Hence, measuring the step length and frequency of an athlete is likely also very important in the context of sprint performance.
Although minimum acceleration magnitude did not emerge as a significant predictor in the regression model, its strong negative correlation with sprint time (r = − 0.803) and strong positive correlation with maximum velocity (r = 0.824) suggests it may represent an athlete’s ability to sustain acceleration and minimize mid-sprint deceleration. This aligns with findings by Nagahara et al. 10, who measured the acceleration profile of 12 elite sprinters across 60 meters using a lab-based environment of 60 infrared cameras. They demonstrated that greater horizontal propulsive impulse and reduced braking forces are essential during the late acceleration phase of the sprint, with vertical force output supporting the maintenance of maximal velocity 10. These observations tie into our findings of higher minimum acceleration magnitude captured from the SportScientia techlayer insole are highly correlated with higher maximal velocity, presumably due to a maintenance of higher forward momentum in faster athletes. In contrast, kinematic variables such as knee angle at full support did not significantly predict sprint time in our model. We believe that kinematic variables such as this while not predictive of sprint performance are more likely related to injury risk while sprinting and may be more valuable in the context of injury risk reduction 11.
Our second regression model (predicting MaxV) further clarified the mechanisms underpinning sprint performance. Maximal velocity was most strongly predicted by maximum step length*step frequency and minimum acceleration magnitude, whereas peak force and time to peak force were not significant contributors. While ground reaction forces are critical to sprinting, our findings align with Colyer et al. 12 and Nagahara & Murata 13, who demonstrate that force orientation (horizontal and vertical) rather than magnitude alone are what differentiate elite sprinters. Hence, it appears that effective sprinters apply force rapidly in a horizontal direction during acceleration phases, and then minimize braking to maximize forward propulsion (i.e. maximal velocity) 13,14.
Regarding the 20 metre findings, the maximal velocity over 20 meters was the strongest predictor of sprint time explaining 74.7% of performance variance. This highlights the importance of achieving high velocities rapidly during the acceleration phase 14. Previous research by Morin et al. 14 demonstrated that elite sprinters’ acceleration capacity is primarily determined by their ability to produce greater net horizontal impulse while minimising braking forces. In part, our results support this concept, as peak force and impulse were moderately correlated with 20 metre time, and the minimum acceleration magnitude, which is a quasi-measure of braking force was also moderately correlated with maximal velocity over 20 meters. As such, interventions targeting performance improvements in 20 metre time should consider measuring maximal velocity across this time given its large contribution to overall time.
While the findings of this study are novel there are some noteworthy limitations. This study is limited by its focus on elite sprinters, which may restrict generalizability. In sub-elite and recreational athletes, it is presumed that variability in the selected measure of our model would be higher, thereby reducing the predictive value. As such, we caution the use of these variables across the full athlete spectrum. In addition, the use of 60 metre sprints excludes the latter deceleration phase often present in 100 metre races. Furthermore, the lack of kinograms to assess the kinematic variables in the 20 metre sprint may strengthen regression models of sprint performance, and thereby influence the model reported in this study may be different when considering kinematics similar to the 60 metre sprint. As a result, future work should develop further comprehensive data related to acceleration cross 20 meters and across the full 100 meters when fatigue among other factors become more relevant.
Conclusion
This study demonstrated that emerging field-based technology can generate highly predictive models of sprint performance in elite athletes. Over 60 metres, performance was predominantly explained by maximal velocity and the product of step length and step frequency, accounting for more than 93% of the variance. Maximal velocity itself was strongly predicted by step length × step frequency and minimum acceleration magnitude, explaining 88.5% of the variance. Across 20 metre, sprint time was almost entirely determined by maximal velocity, while 20 metre maximal velocity was best predicted by the percentage of maximal velocity achieved at 60 metres. These results highlight the critical role of velocity, step kinematics, and acceleration-related metrics in both sprint phases and underscore the value of portable field-based systems for accurate performance profiling.
Declarations
Abbreviations
CT
contact time
IMU
inertial measurement units
MaxV
maximum velocity
MaxV_20m
maximum velocity across 20 metres
MaxStepFreq
maximum step length x maximum step frequency
min_accel_magnitude
minimum acceleration magnitude
peakforce_CT
peak force x contact time
percent_maxV60
percentage of maximum velocity at 20 metres relative to 60 metres
TAV
thigh angular velocity
IC
initial contact
SEE
standard error of the estimate
VLR
vertical loading rate
Ethics approval and consent to participate
A
A
A
All participants provided informed consent in accordance with the Declaration of Helsinki statement and ethics approval of the University of Melbourne by the human ethics research committee (ethics ID 30942).Consent for publication
Not applicable
A
Data Availability
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.
Competing Interests
The authors declare that they have no competing interests" in this section.
Funding
No funding was received from government grants or organizations. All research was funded by the authors business privately.
A
Author Contribution
TS and NC equally shared the workload of this project. XE provided statistical expertise and provided input to the preliminary statistical methods as part of this study. TS handled the ethics application, data analysis and manuscript preparation. NC recruited participants for the study, collected data and assisted with data analysis and manuscript preparation.
Acknowledgements
Not applicable
References
1.
Morin JB, Edouard P, Samozino P. Technical ability of force application as a determinant factor of sprint performance. Med Sci Sports Exerc. 2011;43(9):1680–8.
2.
Rabita G, Dorel S, Slawinski J, et al. Sprint mechanics in world-class athletes: A new insight into the limits of human locomotion. Scand J Med Sci Sports. 2015;25(5):583–94.
3.
Arnason A, Andersen TE, Holme I, Engebretsen L, Bahr R. Prevention of hamstring strains in elite soccer: an intervention study. Scand J Med Sci Sports. 2007;18:40–8. pmid:17355322.
4.
Cross MR, Samozino P, Brughelli M, Morin JB. Methods of power-force-velocity profiling during sprint running: A review. Sports Med. 2017;47(7):1255–69.
5.
Haugen TA, Buchheit M. Sprint running performance monitoring: Methodological and practical considerations. Sports Med. 2016;46(5):641–56. 10.1007/s40279-015-0446-0.
6.
Healy R, Kenny IC, Harrison AJ. Profiling elite male 100 m sprint performance: The role of maximum velocity and relative acceleration. J Sports Health Sci. 2022;11(1):75–84. 10.1016/j.jshs.2019.10.002.
7.
Tam R, Yao W. Biomechanical determinants of sprint performance: A machine learning approach to elite 100-m sprinting. J Biomech. 2024;161:111534. 10.1016/j.jbiomech.2024.111534.
8.
Haugen T, Seiler S, Sandbakk Ø, Tønnessen E. The training and development of elite sprint performance: an integration of scientific and best practice literature. Sports Med - Open. 2019;5(1):44.
9.
Weyand PG, Sternlight DB, Bellizzi MJ, Wright S. Faster top running speeds are achieved with greater ground forces not more rapid leg movements. J Appl Physiol. 2000;89(5):19919.
10.
Nagahara R, Matsubayashi T, Matsuo A, Zushi K. Kinematics of transition during sprint acceleration from a block start. J Sports Sci. 2014;32(7):640–6.
11.
Bramah C, Mendiguchia J, Dos’Santos T, Morin JB. Exploring the role of sprint biomechanics in hamstring strain injuries: a current opinion on existing concepts and evidence. Sports Med. 2024;54(6):783–93.
12.
Colyer SL, Nagahara R, Salo AIT. Kinetic demands of sprinting shift across the acceleration phase: Novel analysis of entire force waveforms. Scan J Med Sci Sports. 2018;28(7):1784–92.
13.
Nagahara R, Murata M, Matsuo A, Kanehisa H, Fukunaga T. Association of sprint performance with ground reaction forces during acceleration and maximal speed phases in a single sprint. J Sports Sci. 2024;42(24):2506–16.
14.
Morin JB, Slawinski J, Dorel S, Saez de Villarreal E, Couturier A, Samozino P, Brughelli M, Rabita G. Acceleration capability in elite sprinters and ground impulse: Push more, brake less? J Biomech. 2015;48(12):3149–54.