Corniani Giulia, Sapienza Stefano, Vergara-Diaz Gloria, Valerio Andrea, Vaziri Ashkan, Bonato Paolo, Wayne Peter
Department of Physical Medicine and Rehabilitation, Harvard Medical School, Spaulding Rehabilitation Hospital, Boston, MA, USA.
Luxembourg Centre for Systems Biomedicine (LCSB), University of Luxembourg, Esch-sur-Alzette, Luxembourg.
Res Sq. 2024 Dec 2:rs.3.rs-5389927. doi: 10.21203/rs.3.rs-5389927/v1.
Tai Chi, an Asian martial art, is renowned for its health benefits, particularly in promoting healthy aging among older adults, improving balance, and reducing fall risk. However, methodological challenges hinder the objective measurement of adherence to and proficiency in performing a training protocol, critical for health outcomes. This study introduces a framework using wearable sensors and machine learning to monitor Tai Chi training adherence and proficiency. Data were collected from 32 participants with inertial measurement units (IMUs) while performing six Tai Chi movements evaluated and scored for adherence and proficiency by experts. Our framework comprises a model for identifying the specific Tai Chi movement being performed and a model to assess performance proficiency, both employing Random Forest algorithms and features from IMU signals. The movement identification model achieved high accuracy (micro F1: 90.05%). Proficiency assessment models also achieved high accuracy (mean micro F1: 78.64%). This study shows the feasibility of using IMUs and machine learning for detailed Tai Chi movement analysis, offering a scalable method for monitoring practice. This approach has the potential to objectively enhance the evaluation of Tai Chi training protocol adherence, learnability, progression in proficiency, and safety in Tai Chi programs, and thus inform training program parameters that are key to achieving optimal clinical outcomes.
太极拳是一种亚洲武术，因其对健康有益而闻名，尤其有助于促进老年人的健康衰老、改善平衡并降低跌倒风险。然而，方法上的挑战阻碍了对训练方案的依从性和执行熟练度进行客观测量，而这对健康结果至关重要。本研究引入了一个使用可穿戴传感器和机器学习来监测太极拳训练依从性和熟练度的框架。在32名参与者进行六项太极拳动作时，使用惯性测量单元（IMU）收集数据，这些动作由专家进行评估并根据依从性和熟练度打分。我们的框架包括一个用于识别正在进行的特定太极拳动作的模型和一个评估表现熟练度的模型，两者均采用随机森林算法和来自IMU信号的特征。动作识别模型取得了较高的准确率（微F1值：90.05%）。熟练度评估模型也取得了较高的准确率（平均微F1值：78.64%）。本研究表明了使用IMU和机器学习进行详细太极拳动作分析的可行性，提供了一种可扩展的监测练习方法。这种方法有可能客观地加强对太极拳训练方案依从性、可学习性、熟练度进展和安全性的评估，从而为实现最佳临床结果的关键训练方案参数提供参考。
