ECSS Paris 2023: OP-AP03
INTRODUCTION: Handball is a high-intensity team sport that requires both physical and cognitive skills. Players must react to constantly changing game situations, executing rapid stop-and-acceleration actions. Reactive agility, the ability to move quickly in response to external cues, is crucial for performance. Developing accessible agility tests and informative outcome measures is essential for understanding athletes´ behavior on the court. In this study, we investigated differences between elite handball players’ planned and reactive agility tasks, focusing on acceleration derived from a single upper-spine-positioned inertial measurement unit (IMU). For a complementary insight into the coordination underlying altered acceleration, we additionally assessed lower extremity joint kinematics. METHODS: Thirty-three elite male handball players from the highest divisions in Germany and Denmark participated. Each player performed five trials of an agility task under planned and reactive conditions. The task required athletes to swipe LED sensors arranged in a trapezoidal setup as quickly as possible. Response times (RTs) were recorded using proximity sensors (Fitlights, Canada). Upper-body acceleration was measured using an upper-spine-mounted IMU (Myomotion/Ultium, Noraxon, USA), focusing on early and later acceleration components. Statistical analyses of RTs and acceleration were conducted using Wilcoxon signed-rank tests. Eight additional IMUs were placed on the feet, shanks, thighs, pelvis, and lower spine to further examine movement patterns and track lower limb kinematics. Statistical non-parametric mapping was used to identify timing and amplitude differences in knee, hip, and foot dorsiflexion between planned and reactive tasks. A significance level of p < .05 was applied to all comparisons. RESULTS: Results indicated significantly slower RTs in reactive tasks compared to planned ones (~200 ms, p <.01). Early movement phases exhibited increased upper-body acceleration in the reactive condition (~9%, p <.01), whereas peak acceleration in later phases remained unchanged. Analyses of joint kinematics revealed delayed movement initiation, more upright starting posture and a greater range of hip and knee flexion in reactive movements (p <.025). CONCLUSION: Our observations suggest that handball players adjust their acceleration strategies in response to cognitive demands. Increased upper-body acceleration in the early movement phase of reactive conditions may reflect compensatory mechanisms due to movement uncertainty . The observed changes in joint kinematics, including a more upright posture and greater range of motion, complement this observation. Together, it could be speculated reactive runs require specific adjustments to body posture, leading to increased upper body acceleration in the early phase of movements. Tracking acceleration metrics via upper-spine IMUs during handball-specific movements provides an accessible and valuable approach for studying movement strategies in athletes.
Read CV Daniel BüchelECSS Paris 2023: OP-AP03
INTRODUCTION: Soccer is a dynamic team sport characterized by high cognitive demand, where quick and efficient decision-making is necessary. Cognitive functions are directly related to how we interact with the environment and make decisions. The present study aimed to identify the key cognitive functions that best predict the performance of youth high-level soccer players on different demands of small-sided games using machine learning algorithms to differentiate the best- and worst-performing players. METHODS: Forty-four male athletes (16.51±0.57 years old) from the U-17 category of two soccer teams participated in the study. For on-field performance evaluation, a 28-round multiple small-sided games protocol was implemented, consisting of 4-minute 3 vs. 3 matches without a goalkeeper. Teams were mixed in every round. On-field performance metrics were: individual goals (IG) as a goal scoring measure, conceded goals (CG) for defensive performance, goals by teammates (GT) as a creativity measure; and net goals (NG) for overall performance. Cognitive flexibility, impulsivity, sustained attention, visuospatial working memory, and tracking capacity were assessed using the Psychology Experiment Building Language, and the Psychophysics software in Python. K-means clustering was applied to segment on field performance metrics into two groups: inferior and superior performance players. For identifying key cognitive functions that best predict different game demands, the five cognitive functions were combined in all possible ways, resulting in 31 datasets used to predict on-field performance levels with seven supervised machine learning algorithms. Grid search and cross validation were implemented for training and testing the algorithms. The 3 best models for each game demand were compared using one-way ANOVA and Tukey’s post-hoc. RESULTS: The standout machine learning models were K-nearest neighbors and neural networks (Multilayer Perceptron). The best models achieved balanced accuracies (BACC) between 69% and 72%, and overall accuracies ranging from 69% to 79%. For predicting IG, cognitive flexibility, and tracking capacity stood out (BACC: 69.6 ± 4.1 %). Sustained attention and visual working memory presented better results in predicting CG (BACC: 72.3 ± 1.1 %). Regarding GT, sustained attention, impulsivity and cognitive flexibility provided best predictions (BACC: 70.5 ± 5.9 %). Lastly, the combination of cognitive flexibility, impulsivity and visual working memory, precisely the executive functions, presented superior results in predicting NG (BACC: 69.3 ± 3.3 %). CONCLUSION: Cognitive functions demonstrated the sensitivity to differentiate athletes performance in small-sided games and highlighted the cognitive functions combinations that better predicted different game demands. These findings demonstrate the potential of cognitive functions evaluation and machine learning application for sports performance analysis, athletes’ selection, and research designs.
Read CV Rafael Luiz Martins MonteiroECSS Paris 2023: OP-AP03
INTRODUCTION: Biological maturation reflects progress toward the adult state, involving structural and functional changes [1]. Growth spurt during adolescence is associated with an increased risk of injuries in youth soccer players. Hence, the monitoring of maturity status and the management of training load of youth players are necessary to avoid overload and injuries. The application of predictive equations to estimate maturity offset (MO) requires manual measurements, which are time-consuming and error-prone if applied at a large scale. Thus, the study aimed at assessing the reliability and validity of a computer vision model to automatically evaluate players standing and sitting height from individual photos, for the determination of the maturity status of elite youth soccer players. METHODS: One hundred forty-six soccer players (age: 9–15 years) from an Italian elite soccer team were measured for standing and sitting height and weight, following ISAK protocols [2], and photographed twice in a standing position. Photos were processed to compute standing and sitting height using a computer vision algorithm (Ultralytics YOLO v9). Mirwald equation [3] was applied to calculate MO using manual heights measures and YOLO-derived heights. The internal consistency reliability of YOLO algorithms for the estimation of heights was assessed using Intraclass Correlation Coefficient (ICC), while absolute reliability was evaluated with coefficient of variation (CV) and standard error of measurement (SEM). Pearson product moment correlation was used to assess the strength of the association between MO calculated with manual heights measures (i.e., the criterion measure) and MO calculated with YOLO-derived heights, as a measure of concurrent validity. Bland-Altman plot was used to visually assess agreement between MO measures together with linear regression analysis for potential bias. RESULTS: YOLO algorithms demonstrated excellent reliability for both standing (ICC=0.961; CV=1.98; SEM=0.531) and sitting (ICC=0.943; CV=3.49; SEM=0.518) height. Pearson correlation coefficient between MO calculated with manual heights measures and MO calculated with YOLO-derived heights showed a strong association (r=0.981; p<0.001). Bland-Altman plot inspection confirmed good data distribution within upper (0.723) and lower (-0.724) limits of agreement. Regression analysis revealed a significant B unstandardized coefficient (B=0.34; p=0.038). CONCLUSION: The YOLO algorithms provided highly reliable and valid values for standing and sitting height, demonstrating a strong consistency and agreement with manual measurement of height. This approach offers a practical, time-efficient, and less invasive method for estimating maturity offset. Finally, a timely evaluation of the maturity status can also optimize the management of training load of soccer players. 1. Malina et al., 2015; 2. International Standards for Anthropometric Assessment, 2001; 3. Mirwald et al., 2002.
Read CV Salvatore MazzeiECSS Paris 2023: OP-AP03