Surface electromyography in personalised modelling of the head and neck

Preoperative estimation of function loss is subjective and unreliable since it depends on the personal expertise of individual physicians. Moreover, each patient is unique and will respond differently to the various treatment options. The Virtual Therapy Group is developing tools to make this tough decision-making process easier, with personalised functional outcome expectations. The idea is to develop a digital doppelgänger. This virtual look-a-like should be able to adapt to individual patients by processing conventional imaging data (magnetic resonance imagining, computed tomography, ultrasound), as well as 3D video measurements to define mobility of anatomical structures, and surface electromyography (sEMG) to account for individual muscle activation patterns. By personalising the currently available generic models with the above data, we could create genuine digital replicas of individual patients.
In clinical practice, some head-and-neck patients regain full function after their treatment and continue their lives with good speech and swallowing function. Others, however, do not and suffer from pathological speech and dysphagia. We think that these differences relate to variation in neural motor control and muscle innervation. Nerve anatomy differs between individuals. Some people may just have more nerve branch innervations for particular muscles than do others.
With sEMG, we can record a crude estimate of muscle activations, which will hopefully enable us to map neural motor commands. This dissertation demonstrates in Chapter 3 that with features extracted from sEMG signals, we can accurately estimate 3D static lip shapes. This promising finding shows that sEMG signals can provide sufficient information on motor control. Chapter 4 demonstrates that a statistical model can adequately predict dynamic movements – visemes (groups of speech sounds that visually look the same), facial expressions, and asymmetric movements – with signals measured from 16 facial muscles. Chapter 5 describes the step from statistical models towards biomechanical models that implement real physics. These models will be advantageous because they follow physical laws and preserve real anatomy and geometry.
In Chapter 6, we elaborate on the process of inverse modelling: calculating the input of muscle activations needed to generate specific functional outcomes – in our case, the 3D lip movements of functions such as speech. Unfortunately, this is a rather complicated procedure, and because of the aforementioned redundancy of the musculoskeletal system, it can lead to multiple solutions. However, we also demonstrate in this chapter that with sEMG we can reduce the solution-space and acquire more patient-specific data on muscle activation. Chapter 7 presents a technical elaboration on inverse modelling, investigating static and dynamic optimisation techniques with and without sEMG. Chapter 8 discusses the work and proposes future research directions on the basis of four main pillars in personalising the generic models.
To conclude, forward modelling will be elementary for driving the model with surgical adaptations and patient-specific learnt muscle-activation strategies, so it could show us the treatment effects directly after surgery. Inverse modelling, on the other hand, could show us any potential compensatory mechanisms, which may differ from patient to patient. Some patients will be able to relearn functions; others will not. With a fully operative digital doppelgänger, clinicians will be able to perform various treatment strategies and compare treatment outcomes at the multidisciplinary meeting to agree upon the best individual treatment strategies. The doppelgänger will also be helpful during counselling, to simulate the functional sequelae of treatment and to better prepare and inform the patient.