Lennartz R, Khassetarash A, Spyrou E, Hallihan A, Eskofier B, Nigg B (2023)
Publication Language: English
Publication Type: Conference contribution, Abstract of a poster
Publication year: 2023
Pages Range: 665
Conference Proceedings Title: ISB Program & Abstract Book
Understanding an ice hockey player’s movements and the restrictions incurred by the protective equipment is crucial for improving the equipment and subsequently, player’s performance. The optimum design of the protective equiment is specially challenging given the complexity of the movements and manuevers in ice hockey. This complexity arises from the multitude of possible variables that describe player’s motion and therefore complex analysis methods are required to help direct the researcher’s attention toward the right variables. The purpose of this work was to utilize artificial neural network (ANN) and layer-wise relevance propagation (LRP) [1] to understand how complex drills in ice hockey were affected by the presence of protective equipment.
Seventeen male ice hockey players (age: 26.4 ± 7.0 years; mass: 86.4 ± 6.5 kg; height:183.8 ± 6.5 cm) skated the long axis of the ice rink as fast as possible. The sprint time over 30 meters was recorded (Brower TCi Timing system) and the movement data was captured by inertial measurements units integrated into the Xsens MVN Awinda System. The sprints were performed twice in either with (Equipment) or without protective equipment (No Equipment) conditions in a randomized order. Individual strides were defined from one ice contact to the subsequent ice contact of the same foot. A total of 24 strides were extracted for each of the participants. The trajectories of 12 joint angles (Figure 1) were extracted and normalized. All trajectories for one movement were concatenated into one single feature vector. A shallow ANN was trained to distinguish whether the sprint stride was performed in Equipment or No Equipment condition. Using LRP, the contribution of each variable to the classification result of the ANN was determined.
On average, the participants performed the sprint drill 1.64 % faster in No Equipment compared to Equipment condition (p = 0.0037). The model was trained 17 times while leaving one participant out each time for testing, reaching an average accuracy of 99.3 %. The average relevance scores that were derived from the ANN model are depicted in Figure 1. Each row corresponds to one rotational degree of freedom, while each column depicts one percent of the stride cycle. The histograms at the top and right part of the figure show the vertical, and horizontal summation of the heatmap respectively. The results show that the ANN can distinguish between the two conditions while observing differences in performance. Thus, it appears that protective equipment impairs performance. The presented data indicate that rotations around the medial-lateral axis (in sagittal plane) and movements associated to the shoulder, knee, and hip joint contributed the most to the classification result.
The proposed methodology based on ANN and LRP was able to highlight the variables and time points that differ between Equipment and No Equipment conditions. We were able to distinguish the important movements that the protective equipment restricts, and future design can leverage this information. In general, this approach allows researchers to investigate a question from a holistic point of view and therefore make informed decisions in complex, multivariate problems.
[1] Bach S et al. PLOS ONE 10(7), 2015
APA:
Lennartz, R., Khassetarash, A., Spyrou, E., Hallihan, A., Eskofier, B., & Nigg, B. (2023, August). The Influence of Protective Equipment on Performance in Ice Hockey. Poster presentation at XXIX Conference of the International Society of Biomechanics (ISB), Fukuoka, JP.
MLA:
Lennartz, Rebecca, et al. "The Influence of Protective Equipment on Performance in Ice Hockey." Presented at XXIX Conference of the International Society of Biomechanics (ISB), Fukuoka 2023.
BibTeX: Download