運動傷害對於運動員的生涯影響極大，尤其職業運動員一旦傷病，輕者則缺席賽季，嚴重則可能斷送整個個人職業生涯; 而對於職業球團，若選手容易受傷無法出賽將需要更複雜的球員調度，因此也盡力的避免選手傷害的發生。但根據美國職棒大聯盟的統計，球員因傷無法出賽的比例卻逐年升高，其中投手相較於其他守備位置的球員，面臨著更高的上肢部位的運動傷害風險。因此，有效的預測傷害的發生將有助於傷害的避免以及即早的治療，以減少對球員造成之永久傷害以及對球團的損失。而運用統計學、機器學習以及深度學習於傷害預測的相關研究皆提出各不同理論的預測模型，希望在實務上能夠降低損失發生的機會。
本研究以大聯盟選手於2020年的傷兵名單為樣本，選取12名上肢部位受傷投手並擷取其傷害發生前以及未受傷的影像資料，並透過深度學習架構OpenPose將影像資料萃取出人體骨幹，並實驗使用人體骨架動作辨識是否在卷積神經網路模型上得到不錯的傷害預測結果。而實驗結果顯示，在VGG19的模型下預測表現為最佳。相較於超深層卷積神經網絡ResNet-50，VGG19的預測準確率為最高，顯示較深的卷積神經網絡沒有帶來較高的準確率。

Sports injuries have a great impact on the career of athletes, especially for professional athletes. Once the injury caused, the one with the minor injuries may be out of the season, and one with the serious injuries may ruin his/her whole career. For professional teams, if the players are injury-prone, they are constantly having to shuffle their pitching rotations. Therefore, learn how to prevent these injuries to players is an important task, both for players and in coaches. However, according to the statistics, the proportion of players unable to play due to injuries has been increasing year by year. Baseball pitchers have been shown to be at a higher risk for sustaining injuries, especially upper extremity injuries than position players. In related literatures, some statistics, machine learning, and deep learning methods have been used for injury prediction.
This study selects 12 pitchers who suffered from upper extremity injuries from the MLB where were in the injured list of 2020. Particularly, their pre-injury and uninjured video data were captured as the samples and the human skeleton was extracted as the input data by using OpenPose. In addition, three different convolutional neural network (CNN) models were chosen to find out the best one by training human skeleton data. The experimental results show that the best performance is obtained by the VGG19 model. Compared with the ultra-deep convolutional neural network ResNet-50, VGG19 has the highest prediction accuracy, showing that the deeper convolutional neural network does not bring higher accuracy.

[1] Bergstra, J., & Bengio, Y. (2012). Random search for hyper-parameter optimization. Journal of machine learning research, 13(2).
[2] Cao, Z., Hidalgo, G., Simon, T., Wei, S. E., & Sheikh, Y. (2019). OpenPose: realtime multi-person 2D pose estimation using Part Affinity Fields. IEEE transactions on pattern fanalysis and machine intelligence, 43(1), 172-186.
[3] D. Tran, L. Bourdev, R. Fergus, L. Torresani, and M. Paluri, "Learning spatiotemporal features with 3d convolutional networks," in Proceedings of the IEEE international conference on computer vision, 2015, pp. 4489-4497.
[4] Ehrmann, F. E., Duncan, C. S., Sindhusake, D., Franzsen, W. N., & Greene, D. A. (2016). GPS and injury prevention in professional soccer. The Journal of Strength & Conditioning Research, 30(2), 360-367.
[5] Erickson, B. J., Chalmers, P. N., Bush-Joseph, C. A., & Romeo, A. A. (2016). Predicting and preventing injury in Major League Baseball. Am J Orthop, 45(3), 152-156.
[6] Escamilla, R. F., & Andrews, J. R. (2009). Shoulder muscle recruitment patterns and related biomechanics during upper extremity sports. Sports medicine, 39(7), 569-590.
[7] Fleisig, G. S., Barrentine, S. W., Escamilla, R. F., & Andrews, J. R. (1996). Biomechanics of overhand throwing with implications for injuries. Sports medicine, 21(6), 421-437.
[8] He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 770-778).
[9] Ji, S., Xu, W., Yang, M., & Yu, K. (2012). 3D convolutional neural networks for human action recognition. IEEE transactions on pattern analysis and machine intelligence, 35(1), 221-231.
[10] John Mishock, “Baseball Pitching: Signs and Symptoms of Pitcher Fatigue to Reduce Arm Injury”, 2017
https://www.train2playsports.com/baseball-pitching-signs-symptoms-pitcher-fatigue-reduce-arm-injury/ (2017)
[11] Johnson, W. R., Mian, A., Lloyd, D. G., & Alderson, J. A. (2019). On-field player workload exposure and knee injury risk monitoring via deep learning. Journal of biomechanics, 93, 185-193.
[12] Josh Myers, “The Cost of Pitching Injuries”, 2020
https://www.dvsbaseball.com/blog/the-cost-of-pitching-injuries

[13] Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25, 1097-1105.
[14] Newell, A., Yang, K., & Deng, J. (2016, October). Stacked hourglass networks for human pose estimation. In European conference on computer vision (pp. 483-499). Springer, Cham.
[15] Pappas, A. M., Zawacki, R. M., & Sullivan, T. J. (1985). Biomechanics of baseball pitching: a preliminary report. The American journal of sports medicine, 13(4), 216-222.
[16] Peng, X., & Schmid, C. (2015). Encoding feature maps of cnns for action recognition.
[17] Polster, J. M., Bullen, J., Obuchowski, N. A., Bryan, J. A., Soloff, L., & Schickendantz, M. S. (2013). Relationship between humeral torsion and injury in professional baseball pitchers. The American journal of sports medicine, 41(9), 2015-2021.
[18] Simonyan, K., & Zisserman, A. (2014). Very deep convolutional networks for large-scale image recognition. arXiv preprint arXiv:1409.1556.
[19] Song, S., Lan, C., Xing, J., Zeng, W., & Liu, J. (2017, February). An end-to-end spatio-temporal attention model for human action recognition from skeleton data. In Proceedings of the AAAI conference on artificial intelligence (Vol. 31, No. 1).
[20] Victor Roman, “Finding Donors: Classification Project With PySpark”, 2019
https://towardsdatascience.com/finding-donors-classification-project-with-pyspark-485fb3c94e5e#_=_