Background: Sarcopenia is a progressive age-related musculoskeletal disorder. Early identification is critical for effective intervention. However, current criteria are time-consuming and require various equipment, limiting their utility. Therefore, we utilized a single inertial measurement unit (IMU) sensor and machine learning to classify possible sarcopenia. Objects: This study aimed to develop a practical machine learning based classification model for possible sarcopenia. Methods: A total of 57 older adults participated and were classified into possible sarcopenia (n = 20) and non-possible sarcopenia (n = 37) groups based on the Asian Working Group on Sarcopenia 2019 guideline. We extracted the mean and variance from the whole-body IMU sensor data during gait and developed five machine learning classification models. Results: The left lower leg sensor demonstrated the highest classification performance among the whole-body sensors. Using the left lower leg sensor data, the support vector machine yielded an area under the receiver operating characteristic curve (AUROC) of 0.79. Notably, integrating demographic variables with IMU sensor features significantly enhanced the model’s performance, achieving an AUROC of 0.92. Conclusion: This study identified the lower leg as the optimal IMU sensor placement for screening possible sarcopenia. Furthermore, the proposed multimodal model, combining IMU sensor data with demographic information, serves as a highly accurate screening tool for possible sarcopenia. This practical model can help early detection of sarcopenia in community settings.

INTRODUCTION
MATERIALS AND METHODS
    1. Participants
    2. Assessment of Possible Sarcopenia
    3. Inertial Measurement Unit Data Acquisition
    4. Data Analysis
RESULTS
DISCUSSION
CONCLUSIONS
FUNDING
ACKNOWLEDGEMENTS
CONFLICTS OF INTEREST
AUTHOR CONTRIBUTIONS
ORCID
REFERENCES

• Junwoo Park(Injury Prevention and Biomechanics Laboratory, Department of Physical Therapy, Yonsei University, Wonju, Korea)
• Kitaek Lim(Injury Prevention and Biomechanics Laboratory, Department of Physical Therapy, Yonsei University, Wonju, Korea)
• Seyoung Lee(Injury Prevention and Biomechanics Laboratory, Department of Physical Therapy, Yonsei University, Wonju, Korea)
• Jongwon Choi(Injury Prevention and Biomechanics Laboratory, Department of Physical Therapy, Yonsei University, Wonju, Korea)
• Donggeon Kim(Injury Prevention and Biomechanics Laboratory, Department of Physical Therapy, Yonsei University, Wonju, Korea)
• Sungmin Chun(Injury Prevention and Biomechanics Laboratory, Department of Physical Therapy, Yonsei University, Wonju, Korea)
• Soyeon Yoon(Injury Prevention and Biomechanics Laboratory, Department of Physical Therapy, Yonsei University, Wonju, Korea)
• Woochol Joseph Choi(Injury Prevention and Biomechanics Laboratory, Department of Physical Therapy, Yonsei University, Wonju, Korea) Corresponding author