In this project, a data-driven modelling algorithm for learning new dynamical models from experimental magnetoencephalography (MEG) data is introduced. This algorithm provides a contrast to existing hypothesis-driven modelling techniques in neuronal dynamics, and is useful for generating new insights from data when hypotheses for the neural mechanisms underlying a process are not readily available. The algorithm utilises universal differential equations (UDEs), combining white-box modelling with machine learning techniques. The algorithm is applied to a single-subject human MEG dataset to produce an oscillator network model. The model captures the frequency-domain behaviour of and interaction between several brain regions of interest during completion of a working memory (WM) task. The machine learning techniques are used to identify the role of attention mechanisms in these interaction dynamics, providing neuroscientists with data-driven insights into the brain dynamics underlying the attention-WM interplay.

Stroke survivors often exhibit motor impairments, which hinder activities of daily living. While grounded robotic perturbation devices can accurately quantify joint dynamics via system identification, their size and fixed positioning limit functional assessments under realistic conditions. To address this gap, we present the design and pilot evaluation of a novel, wearable perturbation device capable of delivering ungrounded force perturbations to the user’s forearm.

The device uses a linear solenoid actuator housed in a wrist brace to generate short, pulse-type forces, thereby inducing small angular deflections (approximately 1–3°) to the arm. An inertial measurement unit (IMU) placed on the brace tracks the resulting movement, while an accelerometer on the solenoid coil measures the perturbation force. Nine healthy participants performed three tasks—relax, resist, and move. Random pulse signals were used to prevent anticipation of the perturbations. The device successfully deflected the arm in all tasks. The largest deflections was recorded during the relax task and smaller, though still measurable, deflections in the resist and move tasks.

Estimated stiffness values in each task indicated that the device could distinguish different levels of joint rigidity, although comparisons with established literature showed some over- or underestimation. Factors such as non-rigid brace attachment and off-center actuator placement contributed to these discrepancies. Despite these limitations, the prototype demonstrates the feasibility of wearable, ungrounded force perturbations for assessing elbow dynamics. Future work will focus on improving the device’s rigidity, exploring multi-degree-of-freedom perturbations, and refining stiffness estimation algorithms to better capture realistic joint behaviors.

Running is one of the most practiced sports worldwide, offering numerous health benefits, but also carrying a risk of injury, mainly at the knee and ankle joints. The origin of running injuries is not fully understood. With predictive neuromusculoskeletal simulations, more insight could be gained into the biomechanical mechanisms that may lead to injuries.
However, in predictive simulations of gait, hyperextension of the knee during stance phase is often encountered. This limits their applicability in research into running-related injuries. It is unclear what causes these unrealistic kinematics, with various studies coming to conflicting conclusions.
This study aims to identify the cause of knee hyperextension in predictive models of running and subsequently, to determine the essential modeling elements for accurately simulating stance knee flexion.

A structured analysis was conducted to investigate the potential impact of the model components within the predictive simulation framework. This framework was divided into four main categories: the objective function, the musculoskeletal (MSK) model, the foot contact model, and the controller. The analysis resulted in numerous hypotheses regarding the element that might be responsible for the simulation of realistic knee kinematics. SCONE, an open-source package for neuromusculoskeletal predictive simulation, was used to test the effect of each hypothesis on the simulated running kinematics. The simulation outcomes were compared to experimental data to assess possible improvements.

The results demonstrate that, in contrast to previous literature, adaptations to the objective function, the MSK model, and the foot contact model have negligible effects on predicted running kinematics. This leads to the conclusion that the controller is essential to focus on when improving knee kinematics. Due to time constraints, multiphase control could not be implemented. Therefore, the exact reflex pathways and phase transitions should be further investigated for the predictive simulation of running before implementation is possible.

Introduction
Chronic pain is an increasing problem in terms of prevalence and disease-related costs. Due to its complexity, it is difficult to treat. Spinal cord stimulation (SCS) is a neurostimulation therapy with a relatively good success rate for patients with severe, intractable chronic pain. The mechanisms of action (MOAs) of SCS are considered to rely on spinal and supraspinal mechanisms. It has been suggested that newer SCS paradigms, such as burst SCS, may act through different MOAs than the traditional tonic SCS paradigm. Tonic and burst SCS are both postulated to act on the lateral pain pathway, which is associated with the location and character of a stimulus, whereas burst SCS is postulated to additionally act on the medial pain pathway, which is associated with the emotional/attentional processing of a stimulus. Somatosensory evoked responses (SERs) can be used to evaluate the processing of somatosensory stimuli and may aid in the unraveling of the MOAs of SCS.

Aim
The aim of this thesis is to assess how burst and tonic SCS affect the supraspinal SERs elicited by non-painful transcutaneous electrical stimulation of the tibial nerve as well as of the median nerve. The two distinct SERs are evaluated using magnetoencephalography (MEG).

Methods
26 chronic pain patients treated with SCS underwent MEG sessions after receiving tonic and burst SCS for one week. Four of these patients additionally underwent a MEG session before SCS treatment. During each session, approximately 200 non-painful electrical stimuli were applied to the median nerve as well as to the tibial nerve to elicit SERs. The SERs were compared in various cortical and subcortical regions of interest (ROIs). The following comparisons were made: 1) SERs in chronic pain patients before SCS implantation versus SERs in the same individuals during SCS, 2) SERs elicited by tibial nerve stimulation versus SERs elicited by median nerve stimulation, 3) SERs during tonic SCS versus SERs during burst SCS, and 4) the SERs in four case studies of two good and two poor responders to the tonic and/or burst SCS paradigms.

Results
22 patients were included for analysis. The number of patients varied among comparisons to facilitate within-patient comparisons. The results suggested an inhibitory effect of SCS on the SER elicited by tibial nerve stimulation, whereas the amplitude of the SER elicited by median nerve stimulation tended to increase during SCS. For both the SERs elicited by tibial nerve and by median nerve stimulation, the SER amplitudes were predominantly higher during burst SCS compared to tonic SCS. Differences in SER amplitude that were observed in the case studies did not correlate with pain relief.

Conclusion
The results suggested a spinal MOA of SCS on the SER, however, supraspinal MOAs likely play a role as well. The results did not suggest that burst SCS additionally acts on the emotional/attentional processing compared to tonic SCS. No evidence was found to support a correlation between the effect of SCS on the SER and the effect of SCS on the pain, underscoring the complexity of the relationship between somatosensory processing and pain perception in the context of SCS.

Temporomandibular disorders (TMD) affect about 5-12 percentage of individuals with consequences such as jaw noises, clicking, myofascial pain, discomfort, limited mandibular range of motion and stress. Treatments depend on the cause and extent of the damage and part of the joint or jaw affected. When exact aetiology of TMD is unclear, generic treatments (splint therapy) are offered. Different oral activities performed on a daily-basis result in different loading conditions on the joint, possibly triggering TMD. These need to be investigated to know the usage of the masticatory system and the potential damage, in order to perform specific treatments. Our work aims at developing an online algorithm that can classify oral tasks performed by individuals. It can be used during daytime or overnight’s sleep to see how often different activities are performed by subjects. A 4 stage wavelet decomposition was employed to the signals and then subjected to feature extraction to train a support vector machine algorithm with. The prediction accuracy was found to be 90 percent for a group of selected oral activities (static, jaw opening, chewing and maximal voluntary clenching). The algorithm had about 80 percent prediction accuracy when classifying both functional (chewing, jaw opening and static) and parafunctional activities (grinding, incisal biting, maximal voluntary clenching, protrusion and laterotrusion) together. However, 80 percent accuracy is regarded as a set back due to the lack of more data. On reviewing the recognised activities, further research on any overuse of muscles or loading on the jaw joint during each activity can be conducted to give specific treatment and therapy preventing any deteriorating actions. Thus,the developed classification algorithm works as a prototype for future studies on online recognition of oral activities.

Migraine is associated with brain dysfunction, possibly due to disturbances in the interactions between distributed cortical regions. Detection of these disturbances in the topological organization of the brain’s functional network would contribute to further understanding of migraine pathophysiology. Altered cortical responses to external stimulation of different modalities are observed in migraine patients, also between attacks (in the interictal state). However, it is yet unclear if abnormalities are detectable in the functional network at rest, i.e. without external stimulation. Here, we assessed abnormalities in migraine functional networks on a global and a local level, based on resting state electroencephalography (EEG) data and graph analysis. Scalp-wide (128-channel)
eyes closed EEG was recorded in 18 episodic migraine patients with and without
aura and 15 healthy controls. We calculated functional connectivity based on
coherence and phase-lag index, and performed graph analysis to characterize
network topology. The minimum spanning tree, a subgraph with maximum
functional connectivity, was used for comparison. No significant differences were
found in network topology, nor in functional connectivity strength between groups. These results demonstrate that this type of graph analyses are not sensitive to any possible abnormalities in the interictal migraine functional network in resting state. Brain dysfunction in migraine might occur only on a local level, making EEG-based graph analysis a less suitable technique to uncover such abnormalities.

By monitoring head movement and orientation in space, the vestibular system can evoke appropriate muscle responses in order to maintain standing balance. The present study investigates whether vestibular-evoked muscle responses are dependent on sensory cues of gravity by examining these responses across varying load and gravity conditions. Standing subjects were exposed to a stochastic electrical vestibular stimulus (EVS, ±5 mA, 0-25 Hz) that induced a vestibular error signal, while vertical loading forces or vestibular signals of gravity were independently modified. A backboard structure limited subjects’ whole-body rotation to the sagittal plane which corresponded with the EVS-evoked sway responses in anteroposterior direction, as the subject’s head was rotated in yaw. Vestibular-evoked muscle responses were greatest when sensory cues of gravity matched the expected terrestrial force of gravity, and decreased when these cues were modified. The reduction was largest when both load- and vestibular-related cues of gravity were different from normal. Our results indicate that the vestibular drive for standing balance control is attenuated when sensory cues of gravity are not congruent to normal (i.e. terrestrial) expectations of standing balance and that the degree of attenuation is dependent upon the cumulative incongruency that arises from multiple sensory cues.

Human joint admittance changes with numerous factors constituting the operational point. For large changes of the operational point, joint admittance can be identified using Linear Time-Varying methods on torque and angular position signals measured on human joints. Out of the available methods, the Skirt Decomposition method was selected due to its nonparametric structure and the limited number of a priori assumptions it makes. Its employment on the identification of human joint admittance was completely novel. The method was applied to a simulation model representing joint admittance and on experimental data measured from the wrist joint. In the experiment, the subjects were changing the applied torque to follow a desired trajectory, while the angle of the wrist was perturbed by the manipulator.With a properly designed multisine input, taking into consideration the speed and complexity of the time variation, a variance accounted for (VAF) close to 100 % was obtained in the simulation study on all the tested conditions. From the experiment, it was seen that the contribution of the time variation in the frequency domain was partially masked by the output noise. The noise level could be decreased by lowering the amplitude of the desired torque, and by removing the voluntary torque from the analyzed data. With a desired torque level ranging between 5% and 20%, and considering the bandwidth between 2 Hz and 20 Hz, the mean power of the output residuals in the frequency domain ranged between 16.2 and 27.1 for all the tested conditions. Furthermore, the time-varying dynamics retrieved from the system function showed a clear correlation with the desired torque trajectory.