In this study, we attempted to capture the variations and characteristics of running form using IMU sensors attached to the waist. We utilized both the magnitude and phase of acceleration and angular velocity near the center of mass, and applied FDA to visualize these aspects of the running form. Although previous studies have been traditionally relying on specialized motion analysis equipment and expertise for interpretation of results, attempts have been made in recent years to capture motion using IMU sensors. Studies have attempted to validate the assessment of whole-body biomechanics using partial acceleration data from various body parts, including the hips, back, wrists, and feet [38, 39]. Human movement arises from temporal changes in posture, involving both limbs and the trunk, and manifests as motion within the spatial dimensions of the center of mass. Typically, the center of mass in an upright posture is believed to be located in the lower part of the torso. However, it is important to note that the center of mass is not fixed and can shift with changes in posture.
The clustering of principal component scores derived from multivariate FPCA resulted in the classification of acceleration waveform patterns representing the running form into three distinct clusters. Although principal component scores were plotted in close proximity for each participant, their positions varied. Among the participants who underwent measurements on two occasions, the second set of principal component scores was also classified into the same cluster as the first set and their positions were also close. This suggests that within individual participants, the principal component scores may exhibit relatively similar distributions (see Supplementary Fig.S4). However, the relationship between acceleration waveform patterns representing running form and fatigue did not demonstrate consistent trends or convergence. Subsequently, we computed the average acceleration waveform representing the running form for both the entire participant group and each cluster. The results indicated a similarity in waveform patterns between the entire participant group and Cluster Cls.1, while Clusters Cls.2 and Cls.3 displayed distinct waveform shapes, each exhibiting unique characteristics. It is worth noting that these three clusters were derived from a relatively small sample size, suggesting that a larger-scale investigation in the future may unveil a greater number of distinct running form clusters. Previous studies have proposed various running-form categories, including grounding patterns, running ability, and sex, with a wide range of possible combinations [40–43]. This implies that variations in the running form among runners should be considered. It may be inappropriate to discuss current performance or future directions simply by comparing them to benchmarks, such as elite runners or group averages.
Moreover, it is crucial to note that this study specifically focused on a group of 17-year-old male athletes engaged in the same track-and-field discipline and training environment. Consequently, subtle differences between the participants are visualized. Similar to the characteristics of conventional principal component analysis, the features of the evaluated dataset can influence the principal component functions obtained from functional principal component analysis. For example, if participant data were assessed alongside elite athletes or recreational runners, the differences in acceleration waveforms representing the running forms of the participants in this study may be perceived as relatively limited. This underscores the importance of understanding the features of the dataset being evaluated according to the intended purpose and interpreting the results appropriately. Furthermore, to visualize individual performance and condition management, it is advisable to consider readily-available running-related motion indicators, such as cadence and walking speed. It would be feasible to conduct analyses that incorporate an individual’s past and present data, compare groups and individuals, and target specific individuals.
Detecting fatigue-related changes in running form is critical. Running coaches routinely assess athletes' condition and offer guidance based on their current running form. Simultaneously, many runners seek to improve their performance by referring to their own or others' running forms. Essentially, running form plays a pivotal role in comprehending conditions and enhancing performance. However, Robbie et al. asserted that even with the trained eye of a running coach, visually classifying running economy by long-distance running coaches remains a challenging task, and assessment of running form is not straightforward [44–46]. The findings confirm that the trend of change in principal component scores of FCPA with changes in fatigue is not constant, confirming that it is difficult to fit the trend of change into a standardized pattern (Fig. 3). In this study, two models were constructed: a group model based on fatigue-free data from all participants and an individual model based on fatigue-free data from each participant. The results of functional logistic regression analysis revealed a relationship between the presence or absence of fatigue and running form in various directions, either singularly or in multiples.
The difference between the group and individual model baselines did not consistently yield conclusive results, emphasizing the need for careful interpretation (Fig. 4). The variations in running form based on the presence or absence of fatigue, as observed through comparison with the group model, may have been influenced by individual variations in running form among the participants (see Supplementary Fig.S6). Both univariate and multivariate results frequently demonstrated that the group model outperformed the individual model in terms of deviance-explained% and ROC values. This suggests that, when estimating the presence of fatigue from the running form, it is crucial to recognize the limitations of comparing oneself with others or extrapolating the results to others.
The benefits and challenges of visualizing the running form are explored with the aspiration that all runners, regardless of whether they are elite athletes, students, or recreational runners, can monitor their daily condition and respond appropriately. Recognizing the signs of changes in health conditions or indicators, such as RRIs, and making decisions to adjust training or rest is a complex task [47, 48]. With the methodology used in this study, obtaining results commonly used in motion analysis, such as joint angles or force intensity, is not feasible. However, by utilizing acceleration data from IMU sensors and applying functional data analysis, it is possible to visualize where changes in the running form occur within the running cycle and how they compare with past data. Although this visualized information may seem insignificant, it can serve as a guideline for managing running forms and cultivating interest in one's own condition and health literacy. Furthermore, if visualization using real-world data generated by runners worldwide during their daily runs becomes possible, it could serve as a benchmark for running form management. Consequently, if there are concerns about daily changes in physical condition, taking proactive steps to seek advice from running coaches or healthcare professionals is recommended. In the future, it would be desirable to establish an environment in which all runners can recognize changes in their condition and respond accordingly.
To elucidate the relationship between various factors such as conditions, performance, individual factors of the subjects, footwear, surface, weather, psychological aspects, and acceleration waveform, it is imperative to conduct repeated investigations in actual running environments. This study presents a preliminary exploration of visualization techniques using FDA. Further investigations are warranted to explore the application of this methodology to data from wrist-worn running watches or smartphones as well as devices attached to different body parts. In addition, ensuring the mutual compatibility of data from IMU devices attached to different body segments remains a crucial challenge for future research. The FPCA conducted in this study creates and classifies the principal component functions based on all available information, providing a relative visualization of existing data. Essentially, it is not an absolute evaluation metric, necessitating the collection of information from all subjects for comparison, including data from athletes or Olympians, which serve as ideal benchmarks. New methodologies that provide absolute evaluation metrics for all subjects are desired to enable a straightforward visualization of runners' conditions and performance.