On-Line Recursive Decomposition of Intramuscular EMG Signals Using GPU-Implemented Bayesian Filtering

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%.

Highly Accurate Real-Time Decomposition of Single Channel Intramuscular EMG

“…and its variance denoted as v S n are also provided by the LMS filter. Details and mathematical derivation of this procedure can be found in [21].…”

“…The estimation are mainly based on the last ∞ observations (Y [n − ∞ + 1 : n]) and scenarios (S[n − ∞ + 1 : n]), where ∞ is the maximum window length. More details about the adaptive estimation formulas can be found in [21].…”

“…3) Initialisation: As presented in [21], no MUs are assumed to be active at the beginning of the decomposition. Thus, the set of active MUs indexes A [1] and the sawtooth sequence T [1] are initialized as empty vector.…”

“…Thus, an exhaustive estimation for all scenarios is impossible. Several approaches for path pruning were proposed in [21] to reduce the number of bifurcated scenarios:…”

“…Moreover, a recruitment model for MUs was added [20] so that the algorithm could be adapted to a varying number of active MUs. Recently, this algorithm was implemented into a parallel computation framework [21] and achieved real time decomposition. However, the realtime constraints led to a decrease in accuracy with respect to the offline implementation.…”