Binary Risk vs No-Risk Classification of Load Lifting Activities Using Features Extracted from sEMG Trapezius Muscle

Many work activities may elicit a biomechanical overload. Several studies have reported that work-related exposure, such as lifting loads, are associated with the development of musculoskeletal pathologies. Recently, in the field of physical ergonomics, several quantitative methods have been developed to assess lifting actions and its biomechanical risk. Moreover, several studies have shown that the combined use of artificial intelligence and wearable sensors provides an improvement in biomechanical risk assessment. In the present study, we assessed the feasibility of Machine Learning algorithms to discriminate biomechanical risk classes defined by means of the Revised NIOSH Lifting Equation. Surface electromyography signals were acquired using wearable sensors placed on trapezius muscles during lifting load tasks performed on 10 healthy volunteers. The sEMG signals were processed in order to extract several frequency-domain features to fed Machine Learning algorithms. Interesting results were obtained in terms of evaluation metrics for a binary NO-Risk/Risk classification; specifically, Gradient Boost algorithm reached accuracy and Area under the Receiver operating curve equal to 0.921 and 0.979 respectively. Study results, although preliminary, proved the feasibility of the proposed methodology to assess the biomechanical risk in a quantitative and automatic way.