Anatomy-Based Assessment of Spinal Posture Using IMU Sensors and Machine Learning

Area of Science:

• Biomedical Engineering
• Kinesiology
• Data Science

Background:

• Posture analysis is crucial for identifying health risks.
• Inertial Measurement Units (IMUs) offer a potential method for objective posture assessment.
• Machine learning (ML) can analyze complex datasets to predict health-related outcomes.

Purpose of the Study:

• To define proxy risk labels for posture using IMU-based angle estimates.
• To investigate the predictability of these posture risk labels from demographic, anthropometric, and lifestyle variables.
• To explore the application of ML algorithms in predicting postural deviations.

Main Methods:

• Thirty healthy young adults (18-25 years) participated.
• IMU sensors were placed at key vertebral levels (C1, C7, T5, T12, L5).
• Random Forest (RF) and Artificial Neural Networks (ANN) were used to predict risks like cervical lordosis, thoracic kyphosis, and lumbar lordosis.

Main Results:

• Incorrect postures (desk work, phone use) correlated with increased risk of lordosis and kyphosis.
• Higher daily physical activity was associated with reduced postural deviations.
• ML models achieved balanced accuracies between 0.55-0.82, with RF showing strong performance for cervical lordosis risk.

Conclusions:

• IMU-based posture angle estimates can identify posture-related risk categories.
• ML models demonstrated predictive relationships between posture risks and personal variables.
• Findings suggest potential for IMU-derived labels in preventive health strategies, though further research is needed.