073: EMG-Driven Models in Biomechanics Research

biomechanics electromyography Mar 28, 2023


Electromyography (EMG) is a procedure that offers a glimpse into the language of our muscles. Although the term seems mundanely scientific, it conceals an almost poetic representation of the intricate conversation between nerves and muscles that facilitates our ability to navigate the physical world around us. Muscles are elaborate engines of force and movement intimately connected to the equally complex nervous system (CNS). In the quiet shadows of our bodies, motor neurons transmit electrical impulses, like invisible whispers, to the muscle fibres, instructing them when and how to contract. Therefore, we may consider EMG an attempt to listen in on this conversation, to eavesdrop on the subtle electrical signals that course through our nerves and muscles.

Electrodes are indispensable tools at the heart of EMG and come in several varieties. Some are needles inserted directly into the muscle tissue, allowing researchers to capture the inherent electrical activity. Others are pads placed on the skin above the muscle of interest, which detect the muscle's electrical signals in a less invasive - albeit less precise - measure of the underlying activity. A pattern of electrical activity emerges when the muscle contracts in response to impulses sent by motor neurons. Electrodes capture this pattern and display it on an oscilloscope to reveal an irregular sequence of peaks and troughs. The peak represents the combined firing of a motor neuron and the muscle fibres it controls. This waveform's overall shape and complexity convey valuable information about the muscle's health and functionality.

Researchers and clinicians gain insights into the nature and severity of various neuromuscular disorders by carefully analysing EMG signals. Changes in neuromuscular function manifest as altered EMG patterns, betraying the otherwise invisible effects of these diseases on the body's muscular and nervous systems. However, EMG has its limitations. For example, needle electrodes can be uncomfortable or even painful for the subject. Furthermore, EMG demands substantial proficiency and experience from the practitioner during the procedure and when interpreting results. Nonetheless, EMG is a remarkable technology that provides us with the capacity to decode signals transmitted between nerves and muscles.

EMG occupies a pivotal role in biomechanics research, serving as a vital nexus between the human body and our ever-evolving understanding of its mechanics. In this post, I will explain how researchers harness EMG data to build detailed musculoskeletal models - elegant simulacra that enable us to peer into the depths of muscle activation and predict the subsequent motion of our joints.

EMG in Biomechanics Research

EMG has established itself as an indispensable tool in biomechanics research. Its capacity to measure the delicate symphony of muscle activations allows us to probe the inner workings of the central nervous system's control over our bodily machinery. Through this understanding, biomechanics researchers have used EMG data to serve as control signals to animate musculoskeletal models, which are driven by muscle-tendon unit (MTU) excitations. This, in turn, propels forward dynamic solutions, analogous to the way our bodies respond to neural commands and perform movement tasks in the real world.

A forward dynamic solution (FDS) is a mode of simulation that aims to predict movement and is rooted in a set of initial inputs and the rigid framework of physical laws that govern our universe. These simulations embrace a bottom-up approach, starting with the neural commands that instruct the muscular actuators. Researchers strive to compute the resulting motion and behaviour of the system as though attempting to glimpse the future through the lens of physics and mathematics. In this way, an FDS serves as both an intellectual scaffold and a prism through which we acquire a deeper understanding of the labyrinthine cooperation between force, motion, and the constraints of the natural world.

When we invoke EMG-driven modelling, we find that the neural command is embodied in the amplitude of the EMG signal – a subtle and profound correlation. Moreover, once processed, this signal unveils the magnitude of the muscle activation patterns. However, the quest for knowledge does not end here. Upon solving the enigma of muscle activation, we must confront the challenge of deciphering the muscle forces that arise. This is achieved through the employment of a muscle model – a construct as intricate as the biological system it seeks to emulate.

But, in our pursuit of realism, we encounter a curious paradox. Muscle models that endeavour to capture the physiological aspects of a muscle in its entirety often find themselves ensnared in a web of computational complexity. Attempts to model forces in multiple muscles produce a computational cost that can become a Sisyphean burden - we may undermine the very edifice of understanding we seek to build. Thus, we find ourselves grappling with the delicate balance between pursuing knowledge and the shackles of limitation imposed by the tools at our disposal.

The Hill-type muscle model, a rather phenomenological approach (i.e., one based on observation), eschews the complexity of more sophisticated representations, such as the Huxley and distributed moment models (Huxley, 1957; Gielen et al., 2000). Instead, it casts a discerning eye on the functionality of the muscle, leaving the intricacies of muscle physiology to those who insist on a more granular perspective. This elegant simplicity renders the Hill-type muscle model a favoured companion of EMG-driven modelling. In the Hill-type model, the contractile element of the muscle fibres is placed in series with the viscoelastic component of the tendon, thus enabling researchers to estimate the force generated by the MTU.

The unique capacity for each muscle to generate force effectively is dictated by factors such as optimal fibre length, maximum contraction velocity (the fastest speed the muscle can contract), and pennation angle (the angle at which the muscle fibres are arranged in relation to the tendon). Thus, we must stay mindful of these specific muscle parameters within the Hill-type model when modelling muscle forces. Simplicity has its allure, but one must never lose sight of the subtle intricacies that comprise the human musculoskeletal system. As with so many other domains, the truth lies at the delicate intersection of elegance and complexity.

Regrettably, the forces emanating from the muscle-tendon units reveal themselves to be nonlinear because they are subject to erratic fluctuations in muscle fibre length during the course of contraction. Consequently, one must endeavour to ascertain the muscle-tendon force at each time point in order to accurately gauge the muscle forces that unfold throughout the span of the modelled movement. We are thus reminded of the inexorable complexities that lie beneath even the simplest of gestures.

As previously touched upon, the anatomy of the MTU occupies a central role in accurately calculating the moment created about a joint. Accordingly, it becomes imperative to precisely determine these variables before incorporating them into the model. The muscle-tendon moment arm is directly related to the calculation of joint moments, but it proves difficult because it perpetually shifts with the joint angle. However, armed with knowledge of the MTU's length and the joint angle, we may triumphantly uncover the moment arm through the deft application of the tendon displacement method (An, 2007). With MTU forces and moment arms in hand, we can then proceed to calculate joint moments by multiplying force and moment arm. Knowing these parameters for every muscle, we are equipped to calculate the net moment of a joint.

The validation of models can be achieved through the employment of a nonlinear optimization technique, one that endeavours to minimize the squared error between the model-predicted joint moment and the measured joint moment. Utilising this method, model coefficients can be honed and tailored to individual participants. But we must tread carefully, for in allowing an excess of variables to be altered, we risk producing models that, burdened with a surfeit of varying parameters, fail to exhibit accuracy and predictive power in the face of novel data. Thus, we find ourselves navigating the delicate equilibrium between a model that faithfully replicates the anatomy of the human body and one that, without an overabundance of adjustable parameters, remains steadfast throughout each step of a time series. In this pursuit, we bear witness to the problematic balance between the demands of accuracy and the virtues of simplicity.

In a manner akin to the Hill-type model, musculoskeletal models must strive to faithfully represent the human form without succumbing to the seductive allure of excessive complexity. The EMG-driven modelling approach serves to illuminate our understanding of the efficacy of injury prevention methods, pathophysiology, and motor control training. One of the many virtues of the EMG-driven modelling approach, when juxtaposed with the optimization approach, lies in its capacity to account for the idiosyncrasies of individual CNS control rather than yielding to the temptation of imposing standardised performance criteria. Furthermore, EMG-driven approaches bestow upon us the boon of computational efficiency, sparing us the labour of estimating muscle excitations through dynamic optimization.

Yet, as is so often the case in the pursuit of knowledge, the EMG-driven modelling approach is not without its limitations. Chief among these is its inability to measure muscular activity within the enigmatic depths of our deep musculature. While intramuscular EMG may offer a glimpse into the realm of deep muscle activity, its invasive nature risks distorting the very behaviour we seek to comprehend during dynamic tasks - the needle electrodes cause discomfort or pain that alters muscle recruitment patterns and movement. Moreover, the technical challenges inherent in EMG data acquisition cast a shadow of doubt on the notion of accurately representing CNS control. Nonetheless, the EMG-driven modelling approach reveals itself to be uniquely suited to the study of individuals with disabilities or a history marred by injury. In these cases, the approach allows us to better understand the adaptive movement strategies employed to accommodate the vicissitudes of disability or the lingering spectre of past injuries.


The EMG-driven modelling approach necessitates a series of calibration procedures to determine a range of physiological parameters. These include tendon slack length, optimal fibre length, EMG-to-activation recursive filter coefficients, maximum isometric force, and nonlinear shape factor (Lloyd and Besier, 2003; Buchanan et al., 2004; Manal and Buchanan, 2013). The prediction of muscle forces using the EMG-driven modelling approach requires these calibrated parameters to be integrated into a musculoskeletal model. While this method offers invaluable insights into the mysterious world of muscle dynamics during movement, it is not without its flaws. The complexity of the modelling procedure can burgeon out of control when multiple segments are involved, rendering the process computationally difficult.

In response to this conundrum, a hybrid method has been devised, one that elegantly marries forward and inverse dynamic solutions. (In contrast to forward dynamic solutions, which start with the driving forces and endeavour to foresee the consequent motion, the inverse dynamic solution adopts a decidedly more retrospective approach - it takes the observed motion and external forces as its guiding inputs and works backwards to ascertain the joint torques, muscle forces, and other internal parameters responsible for the genesis of the motion in question.) This union proves advantageous, as it facilitates a cross-validation measure of the joint moments calculated from the forward dynamics method within the error margin of the inverse dynamics method. Thus, we find ourselves poised at the precipice of understanding, forever seeking to illuminate the elaborate interplay between form and function that governs our very existence.

If you found this content interesting, please share it with someone you think might benefit from seeing it. If you want to dive deeper into the topics discussed here, you can find out more about receiving a world-class education in the biomechanics of flexibility and stretching by clicking here.


An, K. N. (2007.) ‘Tendon Excursion and Gliding: Clinical Impacts from Humble Concepts.’ Journal of Biomechanics, volume 40, number 4, pages 713-718.

Buchanan, T. S. et al. (2004.) ‘Neuromusculoskeletal Modelling: Estimation of Muscle Forces and Joint Moments and Movements from Measurements of Neural Command.’ Journal of Applied Biomechanics, volume 20, number 4, pages 367-395.

Gielen, A. W. et al. (2000.) ‘A Finite Element Approach for Skeletal Muscle Using a Distributed Moment Model of Contraction.’ Computer Methods in Biomechanics and Biomedical Engineering, volume 3, number 3, pages 231-244.

Huxley, A. F. (1957.) ‘Muscles Structure and Theories of Contraction.’ Progress in Biophysics and Biophysical Chemistry, volume 7, pages 235-318.

Lloyd, D. G. and Besier, T. F. (2003.) ‘An EMG-Driven Musculoskeletal Model to Estimate Muscle Forces and Knee Joint Moments in Vivo.’ Journal of Biomechanics, volume 36, number 6, pages 765-776.

Mana;, K. and Buchanan, T. S. (2013.) ‘An Electromyogram-Driven Musculoskeletal Model of the Knee to Predict in vivo Joint Contact Forces During Normal and Novel Gait Patterns.’ Journal of Biomechanical Engineering, volume 135, number 2, article 021014.