Scholarlink

<p>Ball tracking and biomechanics technologies are increasingly used in baseball to optimize performance and monitor pitcher health. However, current applications lack the ability to estimate injury-related joint loads on a pitch-by-pitch basis without using motion capture. Peak elbow varus torque is widely recognized as a clinically valid surrogate for injury risk, providing a useful target outcome for predictive modeling. This study developed and validated a machine learning approach to estimate peak elbow varus torque directly from ball flight metrics using data from 143 professional pitchers and 2984 pitches. A random forest model trained with leave-one-subject-out cross-validation predicted peak elbow varus torque with a root mean square error (RMSE) of 3.41 Nm and a coefficient of determination (<i>R</i><sup>2</sup>) of 0.94, substantially outperforming linear regression (RMSE: 12.84 Nm, <i>R</i><sup>2</sup>: 0.05). Limits of agreement (−6.44–6.95 Nm) fell within previously established ranges, distinguishing injured from non-injured pitchers. Permutation-based feature importance analysis identified release speed, spin axis, and release position (vertical and horizontal) as the most important predictors. These findings demonstrate the feasibility of leveraging ball tracking technology as a scalable, non-invasive tool to monitor biomechanical workload, supporting more informed decisions around performance enhancement and injury prevention in baseball.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Predicting peak elbow varus torque from ball tracking release metrics with machine learning in professional baseball pitchers

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月15日

查看全文
  • R. Connor Moore,
  • Brittany Dowling,
  • Jason M. Avedesian,
  • Edward C. Mitchell,
  • Melissa A. Boswell,
  • Reed D. Gurchiek

摘要

Ball tracking and biomechanics technologies are increasingly used in baseball to optimize performance and monitor pitcher health. However, current applications lack the ability to estimate injury-related joint loads on a pitch-by-pitch basis without using motion capture. Peak elbow varus torque is widely recognized as a clinically valid surrogate for injury risk, providing a useful target outcome for predictive modeling. This study developed and validated a machine learning approach to estimate peak elbow varus torque directly from ball flight metrics using data from 143 professional pitchers and 2984 pitches. A random forest model trained with leave-one-subject-out cross-validation predicted peak elbow varus torque with a root mean square error (RMSE) of 3.41 Nm and a coefficient of determination (R2) of 0.94, substantially outperforming linear regression (RMSE: 12.84 Nm, R2: 0.05). Limits of agreement (−6.44–6.95 Nm) fell within previously established ranges, distinguishing injured from non-injured pitchers. Permutation-based feature importance analysis identified release speed, spin axis, and release position (vertical and horizontal) as the most important predictors. These findings demonstrate the feasibility of leveraging ball tracking technology as a scalable, non-invasive tool to monitor biomechanical workload, supporting more informed decisions around performance enhancement and injury prevention in baseball.