The deep landing error scoring system

Our aim was to provide evidence that the Landing Error Scoring System (LESS) – an injury-risk screening tool – can be automated using deep-learning based computer vision combined with machine learning methods, and compare LESS prediction between cropping and machine learning methods. 2D videos from 320 drop-jump landings were analysed in OpenPose and cropped to key frames manually (clinician) and automatically (computer vision). Random forest outperformed linear and dummy regression methods, yielding the lowest mean absolute error (1.23) and highest correlation (r = 0.63) between clinical and automated scores, reaching reasonable sensitivity (0.82) and specificity (0.77) for risk categorisation. Cropping method did not impact LESS prediction.