A deep learning approach for lower back-pain risk prediction during manual lifting

Area of Science:

• Occupational health and safety
• Biomechanical engineering
• Machine learning applications

Background:

• Occupationally-induced back pain significantly impacts industrial productivity and worker well-being.
• Accurate detection of risky lifting behaviors is crucial for injury prevention and reducing associated costs.
• Challenges in recognizing lifting risks include small datasets and subtle sensor data features.

Purpose of the Study:

• To develop and evaluate a novel deep learning method for classifying lifting risk using sensor data.
• To improve the accuracy of identifying incorrect lifting techniques to prevent back injuries.
• To demonstrate the adaptability of deep convolutional neural networks (CNNs) for complex industrial activity classification.

Main Methods:

• A novel deep 2D convolutional neural network (CNN) was developed for classifying lifting biomechanics.
• The method utilized accelerometer and gyroscope data from 10 subjects performing 720 lifting trials.
• No manual feature extraction was required, relying on the CNN's automated feature learning capabilities.

Main Results:

• The proposed deep CNN achieved a classification accuracy of 90.6%.
• This accuracy surpassed that of an alternative CNN and a multilayer perceptron (MLP).
• The deep CNN demonstrated superior performance in classifying subtle lifting risk features.

Conclusions:

• Deep CNNs offer a powerful and accurate approach for classifying lifting risks from sensor data.
• This method can help reduce occupational back injuries and associated economic losses in industry.
• The deep CNN framework is adaptable for classifying other complex industrial activities.