Core lifting task assessment using time-frequency distribution of surface electromyogram signal

Manual material handling (MMH) is commonly practised in the majority of industrial working environments. However, prolonged and incorrect MMH can cause fatigue, resulting in musculoskeletal disorders (MSDs). Workers who have suffered and fully recovered from MSDs following treatment and rehabilitation, are constantly evaluated to determine their residual functional abilities. However, the functional capacity evaluation (FCE) presently in use to measure a person’s physical ability to perform specific work activities depends on the visual observations of a therapist. A crucial constraint inherent in the FCE test is the likelihood that information other than visual observations could influence the therapist's decision. Recent studies indicate that strong characteristics of surface electromyography (SEMG) on muscle performance exist. Therefore, this study has aimed to extend these findings by improving the reliability and validity of the FCE by considering SEMG signals to automatically determine the work level categories of individuals. Eleven healthy control subjects without a previous history of MSD and eleven validation subjects with a previous history of MSD participated in an experiment in performing the FCE’s core-lifting task. Surface EMG signals were collected from four muscles; right and left biceps brachii (BB), and the right and left erector spinae (ES). Although given the SEMG signal is a highly complex and non-stationary signal, the timefrequency distribution (TFD) technique was used to automatically segment and process the signal. A new auto-segmentation through a spectrogram was utilised to reduce the computation complexity of processing the long EMG signal recording demonstrating excellent performance regarding accuracy, compared to conventional segmentation techniques. For the processing stage, three TFDs; spectrogram, Gabor transform, and Stockwell transform were tested to determine the best TFD for the pattern recognition system. While Stockwell transform has higher computation complexity, this technique was the best in terms of accuracy. Three parameters were extracted from the surface EMG signals and three new features (muscle strength, muscle power, and muscle endurance) were estimated from the average RMS voltage (Vrms(avg)) which became input to the classifier. A hybrid combination of Linear Discriminant Analysis and Support Vector Machine demonstrated a 96% accuracy of, 100% sensitivity, 92% specificity, 100% precision and 0.0035 crossvalidation error. In conclusion, this study demonstrated that the new EMG-based FCE was able to analyse the subject’s performance, work level categories and automatically classifying these, thereby, lessening the possibility of error caused by the therapist.

Text FK 2019 14 ir.pdf Download (1MB)

Actions (login required)

View Item