No abstract

Real-time intramuscular electromyography (iEMG) decomposition, which is largely required in the neurological studies and applications, is a complex procedure that involves identifying the motor neuron spike trains from a streaming iEMG recording. We have previously proposed a sequential decomposition algorithm based on a Hidden Markov Model of EMG, that used Bayesian filter to estimate unknown parameters of motor units (MUs) spike trains, as well as their action potentials (MUAPs). In this paper we present a parallel computation implementation of this algorithm on Graphics Processing Unit (GPU), as well as a number of modifications applied to the original model in order to achieve a real-time performance of the algorithm. Specifically, the Kalman filter, previously used to estimate the MUAPs, is replaced by a least-mean-square filter. Additionally, we introduce a number of heuristics that help to omit the most improbable decomposition scenarios while searching for the best solution. Then, a GPU-implementation of the proposed algorithm is presented. Dozens of simulated iEMG signals containing up to 10 active MUs, as well as five experimental fine-wire iEMG signals acquired from tibialis anterior, were decomposed in real time. The accuracy of decompositions depended on the level of muscle activation, but in all cases exceeded 85%.

This paper describes a sequential decomposition algorithm for single channel intramuscular electromyography (iEMG) generated by a varying number of active motor neurons. As in previous work, we establish a Hidden Markov Model of iEMG, in which each motor neuron spike train is modeled as a renewal process with inter-spike intervals following a discrete Weibull law and motor unit action potentials are modeled as impulse responses of linear time-invariant systems with known prior. We then expand this model by introducing an activation vector associated to the state vector of the Hidden Markov Model. This activation vector represents recruitment/derecruitment of motor units and is estimated together with the state vector using Bayesian filtering. Non-stationarity of the model parameters is addressed by means of a sliding window approach, thus making the algorithm adaptive to variations in contraction force and motor unit action potential waveforms. The algorithm was validated using simulated and experimental iEMG signals with varying number of active motor units. The experimental signals were acquired from the tibialis anterior and abductor digiti minimi muscles by fine wire and needle electrodes. The decomposition accuracy in both simulated and experimental signals exceeded 90% and the recruitment/derecruitment was successfully tracked by the algorithm. Because of its parallel structure, this algorithm can be efficiently accelerated, which lays the basis for its future real-time applications in human-machine interfaces, e.g. for prosthetic control.