Predicting Wrist Posture during Occupational Tasks Using Inertial Sensors and Convolutional Neural Networks.

Area of Science:

• Ergonomics and Occupational Health
• Biomedical Engineering
• Machine Learning Applications

Background:

• Traditional ergonomic assessments rely on time-consuming video analysis for estimating wrist postures.
• Wearable sensors and machine learning offer potential for automating posture assessment, increasing data availability for research and clinical practice.

Purpose of the Study:

• To develop and validate a novel method for predicting wrist posture using inertial measurement units (IMUs) and a deep convolutional neural network (CNN).
• To quantify the accuracy and reliability of the IMU-based system compared to optoelectronic motion capture (a gold standard).

Main Methods:

• Ten participants performed simulated occupational tasks while wearing IMUs on the wrist and hand.
• A deep CNN was trained using data from IMUs to classify wrist posture in flexion/extension and radial/ulnar deviation.
• The model was evaluated using a leave-one-out cross-validation approach.

Main Results:

• The system achieved 65% agreement (κ = 0.41) for wrist flexion/extension and 60% agreement (κ = 0.48) for radial/ulnar deviation compared to optoelectronic motion capture.
• The prediction accuracy and reliability are congruent with published values for human estimators.
• The automated system significantly reduces the time required for postural assessment.

Conclusions:

• The developed IMU-based CNN system provides an accurate and reliable method for estimating wrist posture during occupational tasks.
• This automated approach can streamline ergonomic assessments, saving considerable time and effort.
• The technology has the potential to enhance the capabilities of practitioners by automating tedious manual postural analysis.