Article Text
Abstract
Background Water polo had the highest injury incidence of all team sports at the 2016 Rio Olympics. Injuries have been linked to external load in training and competition (Mountjoy et al, 2015, Wheeler et al, 2013). External load monitoring is likely instructive in managing loads to optimise performance while minimising injury risk. Inertial measurement units and machine learning techniques have shown promise in other sports in monitory external load.
Objective To investigate the performance of a decision tree model using peak resultant acceleration (RAccPEAK)and angular velocity (AngVelPEAK) variables during water polo activities - high intensity throwing (HIT), low intensity throwing (LIT), blocking with ball contact (BWBC) and swimming (SWIM) compared to video analysis.
Design A cross-sectional study
Setting A sports institute pool.
Participants We recruited ten elite female water polo players – 21.2 (SD 4.8) years old, 8.7 (SD 4.6) training years.
Interventions Two Blue TridentTM inertial measurement units were applied to each athlete to collect kinetic variables during a standardised baseline test. Each test was recorded using a digital video recorder and coded for activity verification. R-Studio was used to analyse the data and calculate the predicted volume of each activity compared to the coded video verification.
Main Outcome Measurements RAccPEAK and AngVelPEAK values in abovementioned activities and model accuracy.
Results Each of the activities showed distinct bandwidths of RAccPEAK and AngVelPEAK. The model recorded the following activity volume (N) with 6 false positive errors with a 96% accuracy – HIT (32) LIT (34), BWBC (22) and SWIM (70).
Conclusions This method shows potential to identify different upper limb activities in water polo. These activities most likely load the upper limb differentially and would be critical to measure in monitoring external training load. External load measurement may assist in optimising training planning.