The influence of intrinsic properties (mass, stiffness, and alignment) on ankle joint dynamics, along with the sensitivity of a parametric optimization model to changes in these properties, was investigated. Experiments were performed on two young adults using a robotic manipulator to apply controlled perturbations comprising multiple ramp-and-holds while recording position, torque, and electromyography (EMG) data.
Position, torque, and EMG data were subsequently used as input for the computational model, producing a modeled torque. The model, designed to parametrize neural and non-neural contributors to joint stiffness associated with spasticity, employs a 23-parameter Hill-type neuromuscular framework to estimate muscle properties by minimizing the difference between modeled and measured torque.
Experimental results revealed that mass and stiffness significantly influenced the required torque, particularly during the hold phases, while alignment had minimal effects. Model sensitivity analysis showed similar trends, with stiffness having the greatest impact on torque, followed by mass. Changes in alignment were minimal, highlighting its redundancy.
Discrepancies were observed between model predictions and experimental data in capturing the effects of mass and stiffness changes. The model predicted equivalent mass effects regardless of angle, whereas experimental data exhibited angle dependency. Furthermore, modifying stiffness did not affect the dynamic phases of the modeled torque but did influence the dynamic phases of the measured torque.
These findings highlight the complex interplay between biomechanical factors and torque generation. Future work should refine the model to capture physical nuances better, improve experimental setups to control joint angles and optimize model parameters in a staged approach. Such efforts will advance the development of the model and deepen the understanding of spasticity, paving the way for more accurate and personalized treatment strategies.

Parkinson's Disease (PD) is a neurodegenerative disorder with four cardinal motor symptoms: bradykinesia, tremor, rigidity, and postural instability. Adaptive Deep Brain Stimulation (aDBS) is a promising treatment for PD that provides stimulation based on the expression of PD symptoms, improving effectiveness and reducing side effects compared to continuous DBS. Smartwatches can facilitate aDBS by enabling continuous detection of tremor and bradykinesia. However, for bradykinesia detection, existing studies using smartwatch data from PD patients' natural environment were limited in study duration and sample size. To address this, the current study collected smartwatch data for up to seven months in PD patients' natural environments. From 22 PD patients, the smartwatch data was pre-processed, features were extracted and analysed, and a Variational Autoencoder (VAE) was trained to develop a bradykinesia detection model. However, the VAE could only learn from frequency-domain inputs and not from time-domain input data after current pre-processing methods. Limitations were identified in data quantity, distribution, and quality, including low-frequency artefacts and noise. Despite these limitations, feature analysis indicated that the data set contains valuable information about PD motor symptoms. The results of the feature analysis and VAE training on frequency-domain inputs suggest that, after addressing limitations in data quantity and quality, it could be possible to train the VAE using smartwatch data from PD patients' natural environments. In conclusion, although a bradykinesia model was not successfully developed, this study demonstrated the required steps for training the VAE and laid the groundwork for future studies to develop a bradykinesia detection model.

To have a better understanding of difference in characteristics between various mother wavelets, this paper presents a comprehensive investigation into the performance of three commonly used non-orthogonal mother wavelets, namely Morlet, Paul and DOG, in a wavelet-based system identification approach when used for evaluating joint impedance. This method is further modified to make the estimation result much closer to the realistic result. Additionally, the optimization of smoothing parameters is explored across ten distinct situations, encompassing diverse stiffness waveforms such as step, square, sine, triangle, and sawtooth, as well as two different input perturbations. Performance metrics, including running time, random error, bias error, total error, and variance accounted for (VAF), are used to assess the performance of the system identification method in each scenario. The result shows that Paul wavelet yields a better result of stiffness estimation together with bias error for most situations after averaging. The DOG has the shortest running time and Morlet wavelet gives the highest VAF and lowest random and total error. The findings of this study contribute to a better understanding of the strengths and weaknesses of various mother wavelets in joint impedance estimation, providing valuable insights for future applications in the field of system identification and parameter estimation in neuromechanics control.

Introduction: Over the past two decades deep brain stimulation (DBS) has emerged as an important therapeutic option for Parkinson’s disease (PD). However, the current DBS programming method, monopolar review (MPR), is time-consuming, requires highly trained personnel and causes discomfort for patients. This study aimed to predict the optimal stimulation contact(s) based on local field potential (LFP) recordings by the implanted leads and a sensing enabled DBS system, and as such, improve the efficiency of DBS programming in PD patients.

Methods: Level-based LFP recordings (OFF-medication) within the first two post-operative weeks in PD patients implanted with directional Sensight leads® and the Percept PC® neurostimulator in the Haga Teaching Hospital were retrospectively analysed. Time and frequency domain data were inspected for artefacts. From the individual theta (4-7 Hz), alpha (8-12 Hz), beta (13-35Hz) and gamma (≥36 Hz) bands the maximum power (Max.) and area under the curve above 1/frequency (AUC_flat) were extracted. The clinically chosen contact during MPR served as reference for all predictions. Machine learning models using AUC_flat features from frequency band combinations were evaluated using nested cross-validation. Two custom ranking methods, pattern based and decision tree, were developed for both beta band features individually. The predictive accuracy (Acc.) of the 1st and 2nd prediction combined was evaluated on a training and unseen test set, considering all data and subgroups based on amount of symptoms during MPR and amount of beta activity above 1/frequency. The ranking methods were additionally compared to an existing algorithm (DETEC).

Results: Recordings from 34 patients (68 subthalamic nuclei) were analysed. Artefacts did not overlap with frequencies of interest or were sporadic of nature. The machine learning model with the highest performance was a linear discriminant analysis combining raw beta and alpha features (AUC: Design: 0.86, Test: 0.69). For the 1st and 2nd predicted contacts combined, the two best performing ranking models were pattern based using AUC_flat (Acc. training set: 86.2%; test set: 100%) and decision tree using Max. (Acc. training set: 87.9%; test set: 100%). Correct pattern based (AUC_flat) predictions were more often 1st opposed to 2nd predictions than correct decision tree (Max.) predictions (Tr = 55.2%, T = 90% vs. Tr = 10.3%, T = 10%). Acc. obtained for subgroups based on amount of symptoms during MPR and amount of beta activity above 1/frequency were similar across all subgroups for the custom ranking methods. The Acc. of the DETEC algorithm was inferior to all custom ranking methods.

Conclusions: This study demonstrates the feasibility of using level-based LFP recordings to predict the optimal stimulation contact in patients with PD. The best results were obtained using the pattern based (AUC_flat) and decision tree (Max.) custom ranking methods. For clinical implementation the decision tree (Max.) ranking method is expected to be favoured. Although prospective research is required to identify the true Acc. of the models in clinical practice, these results show potential to halve the required DBS programming time (only two out of four contacts require evaluation), and can thus improve DBS programming efficiency.

Patients with neuromuscular diseases that are unable to speak, but whose cognitive ability has been maintained, can be benefited from Brain Computer Interfaces (BCIs). The decoding of inner (covert) speech from EEGs consists of one of the state of the art methods that aim to tackle this issue. High variability between subjects, as well as low signal to noise ratio (SNR) undermine the methods used, and introduce the need for computer assisted solutions. Thus, machine learning models as well as large amounts of recorded data are required to design effective algorithms and produce substantial results. In this study, covert vowel classification was performed in a systematic way, by making use of two openly shared databases from literature; the Coretto database, that contains EEG recordings of native Spanish speakers, and the DAIS dataset, which includes EEG recordings of native Dutch speakers. Six classifiers were initially selected to perform 5-class classification: a Random Forest (RF), a k Nearest Neighbours (kNN), a Gaussian Naive Bayers (GNB), a Deep Convolutional Neural Network (DCNN), a Shallow Convolutional Neural Network (SCNN) and a Long Short Term Memory Recurrent Neural Network (LSTM). The DCNN outperformed the other methods, with average intra-subject accuracies of 35% for Coretto and 39% for DAIS (chance level 20%). Afterwards, an Overt versus Covert trials experiment was implemented, to test the limits of overt speech decoding from EEGs. The overt result was slightly higher than covert, with an intra-subject average value of 37.8% for Coretto and 40.5% for DAIS (chance level 20%). Finally, binary classification was performed to identify those pairs of vowels that can be classified more efficiently. Vowels /a/ and /u/ seemed to perform better in average in both datasets (average of 64.8% for Coretto and 64.4% for DAIS with a chance accuracy of 50%). Future work should focus on identifying the useful parts of the EEG recordings, increasing the SNR and the resolution of the electrodes, and defining the most appropriate dictionaries of words/vowels for a BCI. Also, more studies should follow systematic ways of comparisons between datasets, to obtain less ambiguous insights and lead this field to improvements.

Epilepsy has been reported in 10-40% of children in the paediatric intensive care unit (PICU). Amplitude-integrated electroencephalography (aEEG), often used as neuromonitoring in the PICU, has some limitations and as a result, caretakers in the PICU may find it challenging to interpret aEEG. This may lead to uncertainty during diagnosis and increases the need for the assistance of the neurophysiology department, a costly EEG and a late diagnosis which might result in irreversible harm.
The aim of this exploratory study was to identify and evaluate observer-based features as stand-alone classifiers and combined in a random forest classifier, that could aid in the accurate classification of seizures in paediatric critical care patients with 4-electrode EEG.
Several features achieved an AUC above 0.60, with the lower border of the aEEG signal yielding the highest AUC of 0.79. The best-performing features maintained their classifying ability when tested on a larger, independent dataset. The random forest classifier did not provide better results.
Our results show that various observer-based features can aid in the identification of seizures in 4-electrode EEG. These additional features could increase the accuracy and decrease the delay and uncertainty in diagnosing epilepsy in aEEG by PICU staff, reducing the need for full setup EEG monitoring and improving patient outcomes in paediatric critical care.

Feedback error learning (FEL) is a classical computational model that describes human motor learning. It consists of forward and inverse models representing internal dynamics and environmental disturbances. Such models can be used as controllers that represent the function of the motor cortex. On top of FEL, a model has been built with jointly trained feedforward and feedback controllers using a neural network. The controllers that actuated six muscles driving a two-degrees-of-freedom arm model were trained offline. This model successfully simulated human learning of point-to-point reaching movements in a horizontal plane when it was tasked to adapt itself in a null field (NF) and a velocity-dependent force field (VF). In this study, we further tested this model in the divergent force fields (DF) and a channelled force field (CF) to observe its performance. The comparison between the simulation results and the experimental evidence suggests that this model can predict some of the key features of the learning process, such as the kinematics, the muscle dynamics, and the impedance profiles. The learning decay was hindered when the lateral error was artificially eliminated by the CF, as reported in the literature. Overall, the model could converge towards realistic

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.

Chronic inflammatory demyelinating polyneuropathy (CIDP) is characterized by sensory and motor deficits. Tremor is a frequent and severe symptom in patients suffering from CIDP. There is limited and contradictory data regarding the effect of tremor on treatment of the underlying polyneuropathy, and on treatment specifically aimed at neuropathic tremor. The pathophysiology of the CIDP tremor is unknown, but there are speculations about both a peripheral and central origin. From previous studies, it is seen that if a tremor’s frequency is substantially load-independent, entrained exclusively by comparable driving frequencies, and only minimally phase-reset by torque pulses, it is considered to be primarily of central origin. A wrist perturbation study using a wrist manipulator was conducted on CIDP patients, with tremors (n=8) and without tremors (control, n=8) to address these speculations, and to understand the characteristics of this neuropathic tremor. The perturbation study involved studying the effects of entrainment perturbations, spring loading via changing the virtual spring stiffness, ramp-and-hold stretches to evoke stretch reflexes, along with the estimation of admittance. The entrainment task showed subharmonic and superharmonic peaks of power spectral densities of EMG in
tremor subjects. Compared to controls, the tremor patients had significantly reduced stiffness, which could be attributed by the loss of muscle strength, and thus the tremor advertently affects muscle activation. No significant difference in M1, and M2 amplitude responses, and reflex latencies were found between the tremor and control group.

Introduction: Spinal Cord Stimulation (SCS) is a successful last-resort treatment for chronic pain patients, although its exact mechanisms of action (MOAs) still need to be unraveled. The MOAs of SCS partly rely on spinal mechanisms (gate control theory) and supraspinal mechanisms likely play a role as well. Pain processing involves a complex network of cortical structures and can be modulated. Conditioned pain modulation (CPM) is a measure to describe the modulation of pain perception. CPM relies on the 'pain inhibits pain' theory, where a nociceptive test stimulus (TS) is modulated by applying a nociceptive conditioning stimulus (CS). CPM is often less efficient in chronic pain patients, but this might be improved by effective SCS.
Objective: The aim of this thesis is to assess how effective SCS affects the supraspinal mechanisms of pain modulation in chronic pain patients. The supraspinal mechanisms are evaluated using magnetoencephalography (MEG) to assess the cortical response to TS before, during and after CPM in chronic pain patients treated with SCS.

Methods: Chronic pain patients treated with SCS underwent MEG sessions after receiving tonic and burst SCS for one week. Each session consisted of three CPM blocks: before, during and after CPM. During each CPM block 22 TS were administered accompanied by CS during CPM, after each block the subject reported a subjective pain rating of TS (and CS). The cortical response to TS, measured using MEG, was analyzed in the time and time-frequency (TF) domain using Brainstorm and Matlab software. TF decomposition was computed in several pain related regions of interest using complex Morlet wavelets. We examined how CPM affected event related spectral perturbations, induced by TS, in the alpha (8-12 Hz) and beta (13-30 Hz) frequency ranges. The average cortical response during tonic and burst SCS was evaluated in all subjects, and separately in the five clearest SCS responders (effective) and the five clearest SCS non-responders (non-effective).
Results: 17 subjects were included. On average a decrease in subjective pain ratings of TS was observed during CPM. In the time domain, TS evoked activity in areas related to the sensory-discriminative aspect of pain. In the TF domain, TS induced event related desynchronization (ERD) followed by event related synchronization (ERS) in the alpha and beta frequency ranges in the bilateral sensorimotor cortices. During CPM the power of beta ERS was significantly decreased during burst SCS. A trend was observed towards a decrease in power of beta ERS for the effective burst SCS group, whereas no decrease was observed for the non-effective burst, effective tonic and non-effective tonic SCS groups.
Conclusion: The results suggest that on average, effective burst SCS decreases the power of beta ERS during CPM, this decrease indicates successful modulation of pain. Therefore, I suggest that effective burst SCS is capable of improving the supraspinal mechanisms of pain modulation, whereas effective tonic SCS is not capable of doing so. This suggestion indicates a partially different MOA for tonic and burst SCS. Future studies containing larger group sizes should validate these findings.

To enable communication for patients who have lost the ability to speak due to severe neuromuscular diseases, covert speech based brain-computer interfaces (BCIs) might be used. These system use neural signals arising from covert speech and translate them into text or synthesised speech. Covert speech is imagining to speak without moving any of the articulators and therefore does not rely on actual motor activity. As recognizing covert speech from neural signals is extremely challenging, machine learning algorithms are deployed. To make use of the full potential of machine learning approaches in the field of decoding covert speech and to accommodate real-world deployment of a BCI, a large number of training samples is required to train the networks.
In this study, a novel database is presented containing EEG and audio data from 20 subjects recorded during the covert and overt pronunciation of 15 Dutch prompts. To validate the recorded data, two speaker-independent classification tasks were performed using a ResNet-50 algorithm as classifier with spatial-spectral-temporal features extracted from the EEG signals. The speaker-independent three-class classification of pre-stimulus (rest) trials versus covert speech trials versus overt speech trials obtained an average accuracy of 70.6% and the speaker-independent five-class classification of five covert vowels (“aa”, “ee”, “oo”, “ie”, “oe”) obtained an average accuracy of 19.6%. Even though the five-class classification task did not reach an above chance level accuracy, the high performance reached by the three-class classification task provides support of the existence of discriminative information in the covert speech segments to decode covert speech in the future.
Future research should focus on EMG artifact detection and on determining the performance per subject to improve the dataset. Furthermore, subject normalisation strategies should be investigated to address the challenges of subject-independent covert speech decoding.

Transcranial Direct Current Stimulation (tDCS) is an upcoming therapeutic tool to improve post-stroke upper extremity motor function by modulating cortical excitability. However, research shows mixed findings on tDCS in stroke patients. The functional and structural brain reorganization may explain the mixed findings: (i) The individual motor function, i.e. the target for tDCS, may be relocated to a non-affected brain region, and (ii) the structural characteristics, i.e. conductivity, of the lesion influences the current propagation in the brain.
This study evaluated the effect of individual optimized tDCS configurations for 21 chronic stroke patients and ten healthy controls using model simulations. We optimized tDCS configurations for two individual targets: (i) a functional target based on EEG during a functional hand task and (ii) an anatomical target based on MRI. The individual optimized configurations are compared with conventional configurations targeting the contralateral and ipsilateral primary motor cortex. The configurations are compared based on the normal component of the simulated electric field at both the functional and anatomical targets. Here we consider a negative field strength inhibitory, therefore undesirable.
The functional target was found ipsilateral in eight of the twenty-one stroke subjects, which indicates the relevance of individual functional targets. The conventional configurations simulated for all healthy subjects a positive field strength in the anatomical target. However, our simulations showed that conventional tDCS can result in negative stimulation for seven stroke subjects in the functional target and two stroke subjects in the anatomical target. The simulations may explain the lack of consistent therapeutic effects of conventional tDCS in stroke patients, and highlight the need for individualized tDCS configurations.

Deep brain stimulation (DBS) is used to treat a variety of movement disorders. In current DBS therapy, the stimulation parameters are manually adjusted based on a subjective assessment of patient’s symptoms. State-of-the-art DBS research focuses on recording brain activity to create a patient-specific neuronal profile that correlates with the patient’s symptom-severity. The neuronal profile could then be used as feedback signal for an adaptive DBS system that automatically adjusts the stimulation. The Medtronic PerceptTM is the first commercially available, fully implanted neurostimulator with sensing technology. In this study, the neuronal activity in the form of local field potentials (LFPs) of eleven patients were recorded with Percept during both stimulation off- and on. To obtain a reliable neuronal profile from the LFPs, the signals should be free from electrocardiography (ECG) artefact and should be synchronized with externally recorded data that informs us about the patient’s symptom-severity. A template subtraction method and a method based on independent component analysis were developed to suppress ECG artefact. Both methods were compared to the ECG removal method of the Perceive Toolbox. Three methods of synchronization were explored: externally tapping, using the stimulation artefact, and by ECG cross-correlation. Results show the template subtraction method to most effectively suppress ECG artefacts while preserving LFP frequency content. To save effort but assure accuracy of the synchronization, the ECG cross-correlation method should be applied though validated by confirming accurate alignment of the stimulation artefact. The development and evaluation of the ECG artefact suppression- and external data synchronization methods contribute to research focused on finding a reliable feedback signal for adaptive DBS.

The automotive industry is foreseeing a future where driver and car will form a synergy, by allowing the car to predict the driver’s intent. Nissan has already successfully predicted the intent to make lane changes in a driving simulator with just six electroencephalography (EEG) electrodes on the movement related cortical potential using Machine Learning. This thesis assesses if the prediction can be improved with EEG layouts containing more electrodes. A large data set was recorded with a similar lane change paradigm in a driving simulator as Nissan’s original study. Seven predefined EEG layouts were assessed: including 4, 9, 18, 25, 32, 48 and 64 electrodes. The linear discriminant analysis classifier was used offline to classify two classes: driving straight or steering. Secondly, to assess if a different selection of electrodes could have performed better, partial least squares (PLS) regression was used to select electrodes based on importance. Lastly, it was also assessed if the classifier could predict the direction of lane change as well as intention, in other words, if the data set consisted of three classes: steering left, steering right and driving straight. There was a statistical difference between the performance of the seven layouts with two classes, but from Layout 3 (with 18 electrodes) and up this difference was no longer significant. The selection of electrodes based on PLS did not prove to be better than the predefined layouts. When the data is divided into three classes, for the prediction of direction, similar results were seen: there was a statistical difference between the layouts and Layout 3 proved to be sufficient. Performance improves with more electrodes, but this effect plateaus. Layout 3 contained a satisfactory amount of electrodes, namely 18, to predict lane change intent. Moreover, the direction of the intended lane change can also be predicted sufficiently with this layout.

Objective: Prediction of comatose patients’ neu-rological outcome after cardiac arrest (CA) is essentialto prevent unnecessary continuation of treatment (withexpensive and scarce resources) and prevent emotionalburdens on families. Electroencephalography (EEG) canaid the prediction of outcome after CA. Visual EEGanalysis is subjective, time-consuming, and might missimportant information. Existing EEG-based machinelearning models are more reliable but lack interpretationof temporal characteristics of EEG. Long-short-term-memory networks (LSTMs) do take advantage of suchtemporal characteristics. Therefore, I hypothesized thatan LSTM would outperform time-insensitive models.Methods: Patients’ neurological outcome at six monthsafter CA was classified as good (no/mild neurologicaldamage) or poor (severe neurological damage/vegetativestate/death). Twelve quantitative EEG features wereextracted from five-minute EEG epochs recorded 12(n=78) or 24 (n=176) hours after CA. A time-insensitivebaseline logistic regression model (LR) and an LSTMwere developed and trained. The performance was eval-uated using the area under the receiver operator curve(AUC) and the sensitivity at 100% specificity (SeSp100)for poor outcome prediction. The LSTM was comparedto the current LR and previously published models.Results and conclusion: The LSTM predicted pooroutcome with AUC=0.90 and SeSp100=0.66, meaningit did not significantly improve performance over LR(AUC=0.89, SeSp100=0.67). However, the LSTM andLR outperformed almost all previously reported mod-els, likely due to the features’ high prognostic power.The SeSp100 was even higher for the LSTM (0.79) andLR (0.78) when using only epochs at 12 hours after CA,suggesting that earlier EEG might further improve prog-nostication.

Purpose – The entropy algorithm is a recently developed statistic for quantifying the complexity of time series data. To date, research of biomechanics and motor control discussed whether entropy algorithms could be used as a convenient tool to identify healthy gait function, to evaluate outcomes of physical therapies and to monitor the progression of disease. Here, we show that Sample Entropy (SaEn) is a sensitive measure for exposing complexity changes in human gait function.

Methods – We analyzed signal complexity changes in electromyography (EMG), ground reaction force (GRF) and joint angle (GA) time series data of asymmetrical step tasks. We used the coarse-grained time series method and the SaEn algorithm, to determine the temporal resolution that contained most complex structures per datatype. Subsequently, we analyzed complexity changes with age and with walking velocity in the selected resolution. We analyzed complexity changes with age, since healthy gait function is known to deteriorate with age. In turn, we analyzed complexity changes with walking velocity, since walking velocity is known to alter gait function. Eighteen young (mean age 23.27 +/- 1.79 years) and nineteen old (mean age 66.37 +/- 5.26 years) subjects were analyzed for an equal number of strides, described by an equal number of samples, to account for the SaEn dataset length bias.

Results – Age increased entropy in EMG signals. Consecutively, age decreased GRF entropy in the medial-lateral (ML) component for short steps and increased entropy for long steps. Lastly, age decreased entropy in GA signals. Furthermore, walking velocity decreased entropy in EMG signals. Consecutively, walking velocity increased GRF entropy in anterior-posterior (AP) and vertical (VE) components and decreased entropy in the medial-lateral (ML) component. Lastly, walking velocity increased entropy in GA signals.

Conclusions – We portrayed that EMG, GRF and GA signals of human gait altered in entropy with walking velocity and with age. Therefore, our results demonstrate the feasibility of SaEn to quantify changes in healthy gait function. Additional research should confirm possible future clinical applications for entropy algorithms.

IntroductionThe Cellular Oxygen METabolism (COMET) monitor measures mitochondrial oxygentension (mitoPO2) in vivo at the bedside with delayed fluorescence ofmitochondrial protoporphyrin (PpIX). Blocking microcirculatory oxygen supply bypressing the measurement probe on the skin allows the measurement of mitochondrialoxygen consumption (mitoVO2). Combining the static mitochondrial parameter(i.e. mitoPO2) and the dynamic mitochondrial parameter (i.e. mitoVO2) canprovide useful information about different aspects of the mitochondrialfunction in sepsis. However, it is not yet known how these mitochondrialparameters relate to the traditional systemic hemodynamic parameters, such aslactate and central venous oxygen saturation (ScvO2). The first aim of thisstudy is to develop an analysis method to derive a measure for oxygenconsumption. The second aim of this study is to investigate the associationbetween the macrocirculation and the mitochondrial oxygenation, in particularmitoPO2 and mitoVO2, and to investigate the association between these twomitochondrial parameters in critically ill patients with sepsis.  Methods We performed a prospectiveobservational study in critically ill patients with suspicion of sepsis whowere admitted to the Intensive Care Unit of the Leiden Medical UniversityCenter. A patch containing 5-aminolevulinic (ALA) was placed on the anteriorchest wall of the patient to induce PpIX in the mitochondria. MitoPO2 wasmeasured by the oxygen-dependent fluorescence of mitochondrial PpIX. MitoVO2was determined by a dynamic mitoPO2 measurement (taking up a series of mitoPO2samples) while blocking the microcirculatory oxygen supply by applying pressurewith the measurement probe of the COMET monitor. MitoPO2 samples and bloodsamples were collected at the same predefined time points. A sigmoid functionwas used to fit the raw mitoPO2 signal of the dynamic mitoPO2 measurements. Amixed linear model was used to study the association between themacrocirculation and the mitochondrial oxygenation and the association betweenthe mitochondrial oxygenation and oxygen consumption. For this second model, wedivided the mitoPO2 levels into a low, medium and high mitoPO2 group. Results Atotal of 202 mitoPO2 measurements were performed in 41 critically ill patientswith sepsis. However, due to poor signal quality or insufficient fitperformance, 29 (14%) dynamic mitoPO2 measurements of different patients wereexcluded for further analysis. The results of this study showed that highermitoPO2 levels were significantly associated with lower lactate levels (β = -0.002744, p = 0.023). Moreover,mitoPO2 levels of the low mitoPO2 group were significantly associated withhigher mitoVO2 rates (β= 0.103822, p < 0.001). Interestingly, mitoVO2 could not be induced in 31(15%) dynamic mitoPO2 measurements.   ConclusionThese results show that mitoPO2 levels are associated with lactate levels incritically ill patients with sepsis. Mitochondria with low oxygen levels are instress and will switch to anaerobic metabolism leading to lactate production.Moreover, these mitochondria have no reserves and will directly suffer from theacute cessation of the oxygen supply because these mitochondria consume thepresent oxygen faster compared to the mitochondria with higher mitoPO2 levels.

System identification techniques to analyse movement disorders are in development, but concrete clinical evidence to receive broad support from the clinical world is lacking. This master thesis proposes two ways to accelerate their development as part of the master’s programmes in Mechanical Engineering and Science Communication. The Mechanical Engineering part focuses on the development of system identification techniques themselves. For this study, an ensemble-based algorithm is developed to identify intrinsic and reflexive joint stiffness during voluntary movements. The algorithm combines two methods from previous research: the Parallel-Cascade method to separate intrinsic and reflexive stiffness and an instrumental variable approach to allow subjects to carry out voluntary movements and apply closed-loop identification. A simulation and experimental study are conducted with the human wrist joint in which a sinusoidal angle reference trajectory is followed using multiple realisations of the same trial. To elicit the modulation of intrinsic parameters, a torque field is applied that is linearly dependent on the wrist angle with a greater flexion resulting in a higher torque. In the simulation study, the intrinsic and reflexive pathways are separated and over 98% of the variance is explained. In the experimental study, the modulation of intrinsic stiffness is estimated with an explained variance of over 90%. Intrinsic stiffness is highest during the peaks of the trajectory and lower during the transition phase from one peak to another. However, the torque response of the experimental study shows that no significant extra torque is added during the reflexive EMG peak. The obtained results underline the promising potential of the algorithm and open up new possibilities once the algorithm is tested on an experimental task where more reflexes are elicited. The Science Communication part focuses on developing a shared mental model to transfer knowledge on the processes behind cross-disciplinary collaborations to mechanical engineers, as devoting time and effort to the metacognitive process helps in coping with it. However, most theories on cross-disciplinarity are written from a social sciences paradigm and are inherently difficult to interpret from a mechanical engineering paradigm. An analogy is developed to link concepts from both epistemologies together. This is done by assembling a theoretical framework containing concepts on the processes behind cross-disciplinary collaborations. These concepts are linked to analogous counterparts from mechanical engineering by relying on conceptual similarities between mechanical stiffness and collaborative pliability. As a result, knowledge of cross-disciplinary collaborations can be immersed with existing conceptual knowledge of mechanical stiffness. Collaboration theories become part of the primary discipline instead of being treated as a separate discipline. The model, called the Collaborative Pliability Model, is assessed by a group of six mechanical engineering students. The students considered the model to be an effective tool for learning about cross-disciplinary collaborations. By using the model, they came up with numerous new factors affecting the success of their collaborative process. This gives a promising outlook for the future and opens the door for new applications for developing an analogy used to transfer knowledge between specific target groups.

Muscles consists of multiple motor units (MUs) that produce the force required for our movements. Motor unitproperties are responsible for the relation between neural excitation and muscle force (excitation-force relation). Theforce of the muscle is dictated by the recruitment and the force output of the different motor units. The recruitmentis physiologically determined by the recruitment thresholds of the motor units, where motor units with a higher forceoutput are thought to have a higher recruitment threshold, conform Henneman’s orderly recruitment principle. Thegoal of this study was to predict the effects of the MU-threshold and MU-force output distributions on the excitation-force relation of a muscle. Assuming that a muscle’s excitation-force relation depends on the values of the motor unitproperties present in the muscle, the recruitment threshold and maximum force distributions among the motor unitscan be derived from muscle force recordings. To achieve the goal a computer model was build to simulate motor unitrecruitment and force production of a group of MUs with different (combinations) of MU-threshold and maximum MU-force output distributions. Both MU-threshold distribution and maximum MU-force output distribution of the musclewere mathematically varied between linear and exponential curves. Input of the muscle model was a linear increasingneural activation profile and the force output profile of the muscle was simulated. The resultant MU-activation andforce patterns were compared with published results of simultaneous MU-recruitment and force recordings obtainedfrom literature. The simulation results showed that the excitation-force profile does not show characteristics thatcan be linked to either the MU-threshold distribution or the MU-force output. As a consequence, force recordingsalone are not sufficient to estimate the recruitment threshold and maximum force distributions. However, the use of acomputer model in combination with additional recorded data such as (surface) electromygraphy (EMG) is expectedto show results from which estimating MU properties is possible.