This study addresses online decomposition of high-density surface electromyograms (EMG) in real time. The proposed method is based on the previously published Convolution Kernel Compensation (CKC) technique and shares the same decomposition paradigm, i.e., compensation of motor unit action potentials and direct identification of motor unit (MU) discharges. In contrast to previously published version of CKC, which operates in batch mode and requires  ∼ 10 s of EMG signal, the real-time implementation begins with batch processing of  ∼ 3 s of the EMG signal in the initialization stage and continues on with iterative updating of the estimators of MU discharges as blocks of new EMG samples become available. Its detailed comparison to previously validated batch version of CKC and asymptotically Bayesian optimal linear minimum mean square error (LMMSE) estimator demonstrates high agreement in identified MU discharges among all three techniques. In the case of synthetic surface EMG with 20 dB signal-to-noise ratio, MU discharges were identified with average sensitivity of 98%. In the case of experimental EMG, real-time CKC fully converged after initial 5 s of EMG recordings and real-time and batch CKC agreed on 90% of MU discharges, on average. The real-time CKC identified slightly fewer MUs than its batch version (experimental EMG, 4 MUs versus 5 MUs identified by batch CKC, on average), but required only 0.6 s of processing time on regular personal computer for each second of multichannel surface EMG.

This paper presents the fully automatic identification of motor unit spike trains from high-density surface electromyograms (EMG) in pathological tremor. First, a mathematical derivation is provided to theoretically prove the possibility of decomposing noise-free high-density surface EMG signals into motor unit spike trains with high correlation, which are typical of tremor contractions. Further, the proposed decomposition method is tested on simulated signals with different levels of noise and on experimental signals from 14 tremor-affected patients. In the case of simulated tremor with central frequency ranging from 5 Hz to 11 Hz and signal-to-noise ratio of 20 dB, the method identified ∼8 motor units per contraction with sensitivity in spike timing identification ≥ 95% and false alarm and miss rates ≤ 5%. In experimental signals, the number of identified motor units varied substantially (range 0-21) across patients and contraction types, as expected. The behaviour of the identified motor units was consistent with previous data obtained by intramuscular EMG decomposition. These results demonstrate for the first time the possibility of a fully non-invasive investigation of motor unit behaviour in tremor-affected patients. The method provides a new means for physiological investigations of pathological tremor.

We describe the method for identification of motor unit (MU) firings from high-density surface electromyograms (hdEMG), recorded during repeated dynamic muscle contractions. A new convolutive data model for dynamic hdEMG is presented, along with the pulse-tonoise ratio (PNR) metric for assessment of MU identification accuracy and analysis of the impact of MU action potential (MUAP) changes in dynamic muscle contractions on MU identification. We tested the presented methodology on signals from biceps brachii, vastus lateralis, and rectus famoris muscles, all during different speeds of dynamic contractions. In synthetic signals with excitation levels of 10%, 30% and 50%, and MUAPs experimentally recorded from biceps brachii muscle, the presented method identified 15 ± 1, 18 ± 1, and 20 ± 1 MUs per contraction, respectively, all with average sensitivity and precision >90% and PNR >30 dB. In experimental signals acquired during low force contractions of vastus lateralis and rectus femoris muscle, the method identified 9.4 ± 1.9 and 7.8 ± 1.4 MUs with PNR values of 35.4±3.6 and 34.1±2.7 dB. In comparison with the previously introduced Convolution Kernel Compensation method, the capability of the new method to follow dynamic MUAP changes is confirmed, also in relatively fast muscle contractions.

The pathophysiology of essential tremor (ET), the most common movement disorder, is not fully understood. We investigated which factors determine the variability in the phase difference between neural drives to antagonist muscles, a long-standing observation yet unexplained. We used a computational model to simulate the effects of different levels of voluntary and tremulous synaptic input to antagonistic motoneuron pools on the tremor. We compared these simulations to data from 11 human ET patients. In both analyses, the neural drive to muscle was represented as the pooled spike trains of several motor units, which provides an accurate representation of the common synaptic input to motoneurons. The simulations showed that, for each voluntary input level, the phase difference between neural drives to antagonist muscles is determined by the relative strength of the supraspinal tremor input to the motoneuron pools. In addition, when the supraspinal tremor input to one muscle was weak or absent, Ia afferents provided significant common tremor input due to passive stretch. The simulations predicted that without a voluntary drive (rest tremor) the neural drives would be more likely in phase, while a concurrent voluntary input (postural tremor) would lead more frequently to an out-of-phase pattern. The experimental results matched these predictions, showing a significant change in phase difference between postural and rest tremor. They also indicated that the common tremor input is always shared by the antagonistic motoneuron pools, in agreement with the simulations. Our results highlight that the interplay between supraspinal input and spinal afferents is relevant for tremor generation.

We compared non-negative matrix factorization (NMF) and convolution kernel compensation techniques for high-density electromyogram decomposition. The experimental data were recorded from nine healthy persons during controlled single degree of freedom (DOF) wrist flexion-extension, supination-pronation, and ulnar-radial deviation movements. We assembled the identified motor units and NMF components into three groups. Those active mostly during the first and the second movement direction per DOF were placed in the G1 and G3 groups, respectively. The remaining components were nonspecific for movement direction and were placed in the G2 group. In ulnar and radial deviation, the relative energies of identified cumulative motor unit spike trains (CSTs) and NMF components were similarly distributed among the groups. In other two movement types, the energy of NMF components in the G2 group was significantly larger than the energy of CSTs. We further performed a coherence analysis between CSTs and sums of NMF components in each group. Both decompositions demonstrated a solid match, but only at frequencies <3 Hz. At higher frequencies, the coherence hardly exceeded the value of 0.5. Potential reasons for these discrepancies include the negative impact of motor unit action potential shapes and noise on NMF decomposition.