Estimating voluntary elbow torque from biceps brachii electromyography using a particle filter

• Published in: Biomedical signal processing and control Vol. 66; p. 102475
• Main Authors: Chatfield, Logan T., Pretty, Christopher G., Fortune, Benjamin C., McKenzie, Lachlan R., Whitwham, Guy H., Hayes, Michael P.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.04.2021
• Subjects: Cumulative distribution function (CDF); Electromyography (EMG); Inverse transform sampling; Logistic function; Motion model; Particle filter; Probability density function (PDF); Process noise; Sensor model; Voluntary torque; Logistic function; Probability density function (PDF); Electromyography (EMG); Sensor model; Process noise; Voluntary torque; Cumulative distribution function (CDF); Inverse transform sampling; Particle filter; Motion model
• ISSN: 1746-8094; 1746-8108
• DOI: 10.1016/j.bspc.2021.102475

•Electromyography correlates with voluntary torque during dynamic elbow flexion.•Models with electromyography and sensor measurements approximated torque.•A particle filter combined the models and considered their uncertainties.•The particle filter accurately estimated voluntary torque for ten healthy subjects.•Allows for future research on optimal assist-as-need control. Stroke is one of the leading causes of disability worldwide. Assist-as-need control is desirable as it can optimise the rehabilitation and potentially greatly improve the patient's recovery from stroke. However, to achieve optimal assistance, the voluntary effort a patient applies must be known. To verify the use of a particle filter to accurately estimate in real-time the voluntary torque from electromyography (EMG) for various subjects and movement speeds, accounting for the non-linear and time-varying behaviour of the muscle. Ten healthy subjects performed dynamic elbow flexion at various speeds. The EMG of the biceps brachii and the torque were recorded. A motion model and different sensor models were developed for each data set, and the particle filter was then applied to improve the estimate of voluntary torque. The performance of the particle filter with the different sensor models was analysed. By combining the motion model and sensor model, and considering their uncertainties, the particle filter improved the estimate of voluntary torque with a mean normalised RMS error across all subjects and movement speeds of 6.56%. The particle filter demonstrated the ability to adapt and improve the estimate of voluntary torque, and so it is suitable for understanding the voluntary efforts and capabilities of a healthy subject. The next stages are to conduct a clinical trial to verify the effectiveness of the particle filter for subjects affected by stroke, and to analyse assist-as-need control based on the estimate of voluntary torque with a wide range of subjects.