Author(s): CHAN, Y.H., LIN, W.C.2, WU, K.P.2, LIN, C.Y.3, HUANG, T.S.1, Institution: NATIONAL YANG MING CHIAO TUNG UNIVERSITY , Country: TAIWAN, Abstract-ID: 1307

INTRODUCTION:
Badminton players frequently suffer from patellar tendinopathy (PT). Identifying the underlying risk factors of the injury is essential to mitigate the occurrence or recurrence of PT. Previous studies have not investigated the potential risk factors of PT in badminton players. Furthermore, previous prospective studies on sports injury risk factors have typically involved only a single measurement, ignoring the complex and ever-changing nature of injury risk. Regular measurements and continuous data monitoring are essential to capture changes in the data. Deep learning techniques have been used to analyze time series and process the collected data. The purpose of this prospective study was to explore the risk factors of PT by utilizing deep learning methods in adolescent badminton players. Concurrently, the deep learning analysis findings were compared with those obtained through traditional analytical methods.
METHODS:
Forty-nine adolescent badminton players who exhibited no PT at the beginning of the study were recruited. The prospective study involved regular assessments twice weekly with eight measurements for four months. Measurement items included (1) training load, (2) ankle dorsiflexion range of motion (ROM), (3) thigh muscle length, (4) Y-balance test, (5) countermovement jump, and (6) agility test. The occurrence of PT within four months was evaluated. This study employed the gated recurrent unit (GRU) in deep learning to analyze time series data as a predictive model and employed SHapley Additive Explanation (SHAP) analysis to elucidate the significance of features. In traditional analysis, baseline data was utilized for logistic regression analysis, with the significance level set at 0.05.
RESULTS:
Out of the 49 participants, five players experienced PT. The GRU model had an accuracy of 91.84% of prediction. The top 3 SHAP features were dominant and non-dominant ankle dorsiflexion ROM and the performance of the Y-balance test. The logistic regression analysis did not find any significant risk factors for predicting the development of PT.
CONCLUSION:
This research employed deep learning methods with SHAP analysis to develop a novel predictive model. Our findings showed that essential features of developing PT were bilateral ankle dorsiflexion ROM and the performance of the Y-balance test. There were trends of decreased ankle dorsiflexion ROM and a disparity in the measurements of both legs in the Y-balance test before developing PT. The utilization of deep learning has the potential to improve the efficacy of injury prevention for PT compared to traditional analysis.
References:
1. Kimura, Y., et al., Trunk motion and muscular strength affect knee valgus moment during single-leg landing after overhead stroke in badminton. Br J Sports Med, 2014. 48(7): p. 620-620.
2. Sprague, A.L., et al., Modifiable risk factors for patellar tendinopathy in athletes: a systematic review and meta-analysis. Br J Sports Med, 2018. 52(24): p. 1575-1585.