Human movements are still poorly understood, mainly due to their high complexity and large number of degrees of freedom embedded in neural commands. This shortcoming can be compensated by analyzing electrical activity of muscles at the surface of the skin, so called surface electromyogram (EMG). The latter offers numerous advantages over clinically more established intramuscular investigations, such as repeatability of measurements, patient’s comfort and acceptance of investigation technique, no risk of infection and lower investigation and examination costs.

For these reasons surface EMG has been extensively used in the fields of neuroscience, rehabilitation, pathophysiological investigations, training of the athletes and in man-machine interfacing, especially for control of active prosthesis. The main challenge in all these applications is accurate identification of neural commands out of recorded EMG. Namely, EMG is composed of many action potentials that are contributed by basic functional units of a muscle, so called motor units (MUs). Central nervous system controls muscle force by controlling the number of active MUs and their discharge rates. Although playing a vital role in EMG, the shape of action potentials is completely irrelevant from the central control viewpoint. Moreover, due to the polyphasic MU action potential shape and asynchronous MU activity, the recorded surface EMG appears to be highly interferential and the muscle activations are typically estimated by calculating the EMG energy envelops. This offers limited insight into the neural commands of skeletal muscles, mainly due to sensitivity of EMG signals on the shape of MU action potentials. The latter depends on muscle anatomy, subcutaneous tissue and contraction level and is one of the main reasons for large variability and low repeatability of surface EMG signals reported in the literature.

This negative impact of motor unit action potentials has been largely ignored in the studies of motor behavior, mainly due to the lack of suitable signal processing techniques. Even more, EMG envelopes have been used to establish a widespread theory of muscle synergies. The latter hypothesizes that a shared co-activation of two or more muscles is modulated by a single neural command from CNS. This is believed to simplify the control of human movements.

In this project, we address the aforementioned problem of EMG processing by separating the information on the shape of MU action potentials from the information on MU discharge patterns. We then use this novel methodology to study the muscle excitation primitives in upper limb movements of healthy and hemiparetic subjects. In particular, we challenge the current muscle synergies estimation from both methodological and physiological viewpoints. From methodological viewpoint, it is not clear to what extent the identified muscle activation patterns reflect the common geometrical changes in investigated muscles rather their excitation commands. From the physiologic viewpoint, it has been demonstrated that the residual 10% - 15% portion of the signal variability, not accounted for by current muscle synergy estimations, carries the information that is crucial for reduction of functional movement errors.

In order to demonstrate their robustness and suitability for analysis of central nervous system disorders, the developed techniques will be used to accurately track the pathological variability in excitation of individual upper limb muscles in the hamiparetic patients and to compare this variability to the functional movement errors as assessed by a haptically-controlled UHD rehabilitation robot. This is expected to provide better information support to rehabilitation decisions and, thus, maximize the patient’s rehabilitation potential.

Figure 1: A – Size and shape of motor unit action potential (MUAP) as detected by three pairs of bipolar electrodes mounted on the surface of the skin at two different elbow joint angles. MUAPs of the same motor unit appear much stronger with the elbow fully extended (left) than with the elbow fully flexed (right); B – Simulated effect of MUAP shape change due to elbow flexion (from fully extended elbow on the left to fully flexed elbow on the right) at the constant 30% muscle excitation. Energy envelop of the simulated signal is depicted in black. Depicted EMG signal was acquired away from the innervation zone; C – the same as in B, but with the compensated MUAP shapes (MUAP shapes are algorithmically removed from the acquired EMG signals); D - Discharge rates of five motor units during the concentric and eccentric phase of dynamic biceps brachii contraction in a healthy young volunteer. The impact of MUAP shape variability has been compensated by the EMG decomposition technique and the identified MU discharge rates offer highly accurate insight into the muscle excitation patterns.