BME2022 Paper Submission & Registration
9th Dutch Bio-Medical Engineering Conference





Powered by
© Fyper VOF.
Conference Websites
Go-previous
16:00   Poster session 2
Curcumin-loaded zein nanoparticles potentiate temozolomide-induced antitumor effects in GBM cells and patient-derived GBM neurospheres
Huaiying Zhang, Yutong He, M. Alejandro Reina Mahecha, Dong Liang, Frank A.E Kruyt, Jan A.A.M Kamp, Inge S Zuhorn
Abstract: Temozolomide (TMZ), a DNA alkylating agent, represents the clinic's frontline chemotherapeutic agent for glioblastoma (GBM) treatment. However, the therapeutic efficacy of TMZ remains very limited due to its rapid degradation, short half-life as well as frequent resistance in glioblastoma. In previous work, we developed curcumin-loaded zein nanoparticles that efficiently traversed an in vitro blood-brain barrier (BBB) model, and investigated its effects on the growth, migration, and viability of glioma C6 cells. Here, we investigate whether curcumin-loaded nanoparticles (CUR-NPs) can sensitize human glioblastoma U87 cells and GBM patient-derived neurospheres to TMZ treatment, administered either as free TMZ or as TMZ-cyclodextrin (TMZ@CD) inclusion complexes. CUR-NPs enhanced the inhibitory effects of TMZ and TMZ@CD on the proliferation and migration of U87 cells as well as on the growth and stiffness of U87 spheroids. Flow cytometric and western blot analysis demonstrate that CUR-NPs sensitized TMZ treatment in U87 cells by enhancing apoptosis, which may be attributed to CUR-NPs ability to increase reactive oxygen species (ROS) level significantly. In addition, the proliferation and sphere forming capacity of patient-derived neurospheres were evaluated by cell viability assay and limiting dilution analysis, respectively. We show that CUR-NPs enhanced the inhibitory effects of TMZ and TMZ@CD on the proliferation and sphere forming capacity of GG16 and GSC23 neurospheres, indicating strong potential of the combined treatment to inhibit GBM stem cells. Provided that CUR-NPs and TMZ@CD both efficiently accumulate in GBM tissue in vivo, combined treatment with CUR-NPs and TMZ@CD is expected to improve the efficacy of TMZ in GBM and reduce its side effects in non-target tissues.
Distal vascular effects induced via cuff-based perturbation
Laura Bogatu, Simona Turco, Massimo Mischi, Lars Schmitt, Pierre Woerlee, Erik Bresch, Gerrit J. Noordergraaf, Igor Paulussen, Arthur Bouwman, Hendrikus H.M. Korsten, Jens Muehlsteff
Abstract: In standard critical care practice, cuff sphygmomanometry is widely used for non-invasive, intermittent, blood pressure (BP) measurements. However, cuff devices offer ample possibility of modulating blood flow and pulse propagation along the artery, potentially enabling measurement of several other hemodynamic parameters in addition to BP. We explore underutilized arrangements of sensors involving cuff devices which could be of use in critical care to reveal information on compensatory mechanisms for improved assessment of the hemodynamic status. In our previous work, we analyzed the response of the vasculature to cuff-based perturbations via non-invasive measurement set-ups. Observations were acquired by means of electrocardiogram and photo-plethysmogram [1, 2]. In this study, our aims are to 1) acquire additional insights by means of invasive measurements and 2) based on these insights, further develop non-invasive, cuff-based measurement strategies. Invasive BP experimental data is collected downstream from the cuff in two patients monitored in the operating room. It is found that highly dynamic processes occur in the distal arm during cuff inflation. Mean arterial pressure increases in the distal artery by 20 mmHg, leading to a decrease in pulse transit time by 20 ms. These changes need to be taken into account for correct interpretation of the vascular response to occlusion-based perturbations. Previous characterizations neglected such distal effects. A lumped-parameter model is developed to reproduce the observed behaviors and to provide a possible explanation of the factors that influence the distal arm mechanisms. The observed dynamic processes and model reveal opportunities for measuring non-standard but highly relevant clinical parameters related to pulse arrival time vs. BP calibration, arterial and venous compliance, peripheral resistance, mean systemic filling pressure, and artery-vein interaction.
Epileptic seizure prediction using tensor-based supervised learning
Seline de Rooij, Borbála Hunyadi
Abstract: Epilepsy is one of the most common neurological disorders: it affects almost 1% of the worldwide population. It is defined by the seemingly random occurrence of spontaneous seizures. For about 70% of the patients anti-epileptic drugs provide adequate treatment. The remaining 30%, however, continue to have seizures, which drastically affects their quality of life, as they live in a constant state of uncertainty of when these seizures occur. Reliable methods for seizure detection and prediction would, therefore, have a significant impact on these patients’ lives, enabling the optimization of time-varying therapeutic prevention strategies. Despite ongoing research efforts involving academia and industry in large international collaborations, epileptic seizure detection and especially prediction is still an unsolved problem. The key to the solution may lie within ultralong-term, real-life datasets that are currently being generated using wearable sensors. The most promising datasets include multiple sources of relevant information besides brain activity, such as cardiac activity, motion, or the circadian rhythm. These biomedical datasets are high-dimensional. Conventional solutions artificially segment such high-dimensional data into shorter one- or two-dimensional arrays, causing information loss by destroying correlations between the data. At the same time, these datasets are ever increasing due to the substantially larger recording durations of wearable sensor technology. Tensors (multi-dimensional arrays) are the data structure of choice in artificial intelligence research to exploit the full potential of such large-scale data in a timely manner without loss of accuracy. Therefore, the goal of this PhD research is to find suitable tensor-based methods to: 1. Combine multichannel EEG and other data modalities. 2. Learn the non-linear mapping between the different modalities and the non-linear separation between classes. 3. Forecast seizures and at the same time quantify the uncertainty of a prediction.
Graphene-based neural electrodes to combine imaging with electrophysiology
Nasim Bakhshaee Babaroud, Merlin Palmar, Andrada Iulia Velea, Sebastian Weingartner, Frans Vos, Wouter Serdijn, Sten Vollebregt, Vasiliki Giagka
Abstract: Electrical neural recording and stimulation in combination with methods such as optical imaging, optogenetics, and magnetic resonance imaging (MRI) in a multimodal manner have been recently used to pave the way towards a much deeper understanding of the nervous system. However, conventional noble metal electrodes used for electrical recording and stimulation might produce photo-induced artifacts in simultaneous electrophysiology with optogenetics and cause information loss due to the opaque electrodes in optical imaging and image artifacts in MRI. Therefore, the need to create optically transparent and MRI-compatible electrodes is increasing. Graphene has recently attracted a lot of attention as a transparent conductive material. The most common graphene growth method, chemical vapor deposition (CVD), uses high temperatures (>900 °C). This prevents the direct growth of graphene on a polymeric substrate in flexible implants. Therefore, current state-of-the-art graphene electrode fabrication has been focusing on manual graphene transfer techniques [1-2]. This technique is also employed to stack several monolayer graphene sheets and improve its electrical characteristics [3], however, has reliability issues. In this work, we present CVD-based multilayer graphene electrodes using a transfer-free process [4]. The fabricated electrode shows a low impedance at 1 kHz that is comparable to those of gold and platinum electrodes with the same size and geometry. The charge storage capacity (CSC) calculated based on cyclic voltammetry (CV) measurement and the charge injection capacity (CIC) calculation based on voltage transient (VT) measurement shows comparable results to those of platinum electrode. The graphene electrodes with different thicknesses did not reveal any photo-induced artifact in the power spectrum of the recorded signal after shining light with 470 nm wavelength to the electrode surface. Moreover, this fully transparent electrode did not show any image artifact in the 3 T MRI scanner. The transfer-free multilayer graphene electrodes yield a high CSC and a low impedance and could be used for the neural interfaces and enable multifunctional electrical and optical recording and stimulation and substitute metal electrodes for MRI study of the nervous system. Such graphene-based implants, when complemented with the right combination of transparent polymeric substrates, have the potential to support even long-term studies [5].
Skip-SCSE multi-scale attention and co-learning method for oropharyngeal tumor segmentation on multi-modal PET-CT images
Alessia de Biase, Wei Tang, Nikos Sourlos, Baoqiang Ma, Jiapan Guo, Nanna Maria Sijtsema, Peter van Ooijen
Abstract: One of the primary treatment options for head and neck cancer is (chemo)radiation. Accurate delineation of the contour of the tumors is of great importance in the successful treatment of the tumor and in the prediction of patient outcomes. This work is based on our participation in the HECKTOR 2021 challenge (Andrearczyk & al., 2021) (Oreiller & al., 2021) organized to develop methods for automatic tumor segmentation on PET and CT images of oropharyngeal cancer patients. We investigated different deep learning methods with the purpose of highlighting relevant image and modality related features, to refine the contour of the primary tumor. More specifically, we utilized a Co-learning method (Xue & al., 2021) and a 3D Skip Spatial and Channel Squeeze and Excitation Multi-Scale Attention method (Skip-scSE-M), on the challenge dataset. The Co-learning model uses a V-Net as backbone and it takes as input PET and CT as separate volumes. The Skip-scSE-M model uses a U-Net as backbone, giving smoother and more accurate predictions compared to a normal 3D U-Net, and takes as input PET and CT concatenated in the channel dimension. The challenge dataset (Andrearczyk & al., 2021) contains 325 cases (224 for training and 101 for testing) of aligned PET-CT images from a total of 6 centers. Patients coming from 1 center were included partly in the training and the rest in the test set. Both methods were trained on 3D image patches of size 144 × 144 × 144, centered on the oropharynx. All cases were annotated by experts and only the annotations of the training cases were provided. Both methods showed to be able to detect tumor areas on the two modalities, focusing on different aspects. The Skip-scSE-M model provides a more stable and robust performance compared to the Co-Learning method. The best results achieved on the challenge test set, with our methods, were 0.762 for mean Dice Similarity Score (DSC) and 3.143 median for the 95% Hausdorf Distance (HD95). With our proposed approach we ranked 10th over 42 different teams, where the best results were DSC of 0.778 and 3.088 of HD95.
Three dimensional reconstructed ultrasound for intra-operative resection margin assessment by a motorized transducer
Baris Karakullukcu
Abstract: Context: During oncological surgery, intra-operative assessment of the resection margins is preferred. Three-dimensional (3D) ultrasound (US), as one of the assessment techniques, enables assessment of the resection margins in resected tongue squamous cell carcinoma. The future aim is that the surgeon decides whether to perform an additional resection based on this 3D US assessment. Inaccurate measurements may lead to undesirable over-treatment or under-treatment. Therefore, intra-operative resection margin assessment of tongue carcinoma of the entire specimen should be highly reliable and accurate. Objectives: To assess whether 3D reconstructed US volumes are more accurate when acquired by a motorized US transducer acquisition compared to freehand US acquisition. Method: Two approaches for acquisition and reconstruction were compared in a phantom study. Approach 1 is a freehand US acquisition reconstructed into a 3D volume by Pixel Nearest Neighbor algorithm from CustusX. Approach 2 is a motorized US transducer in a rails acquisition reconstructed by stacking raw 2D US images into a 3D array. A CT scan of the phantom is utilized as a reference model. Quantitative analysis of the image quality after reconstruction is assessed by the SNR, CNR and slope in pixel intensity at a tissue transition. Reproducibility and variability are assessed by performing iterated acquisition for each approach by different operators and evaluated using Mean Square Error or the Structural Similarity Measure. Additionally, the volumes in the phantom were segmented in both the CT and 3D US volumes and compared. Based on these segmentation, the resection margins were computed and compared as well. Results: Ongoing research. Conclusion/Hypothesis: Motorized US acquisition reconstructed by stacking the 2D raw US image performs better than freehand US acquisition reconstructed by the Pixel Nearest Neighbor algorithm based on the evaluation metrics. Also the reproducibility and variability was better in approach 2.
Deep learning predict the changes of pulmonary nodules at follow-up CT
Jingxuan Wang, Peter van Ooijen, Rozemarijn Vliegenthart
Abstract: 1 Introduction and objective The early manifestations of lung cancer are pulmonary nodules, which can be found using chest computed tomography (CT). For clinical diagnosis, patients usually undergo multiple follow-up screenings, which helps radiologist to track changes in nodules. Compared to the baseline screening, radiologist need to determine the disappearance or growth of lung nodules. Even no change is a matter of concern to radiologists. However, repeated CT screening can cause psychological pressure on the patients. Therefore, with the help of deep learning, this research aims to predict the changes of pulmonary nodules, which can help radiologist to reduce the burden of CT reading. More importantly, our research could further reduce the anxiety of patients, avoid unnecessary follow-up examinations and provide advanced diagnosis for early intervention and treatment. 2 Material and method This research will be carried out in Dutch-Belgian Randomized Lung Cancer Screening Trial (NELSON). In the NELSON, we selected 750 participants with 964 indeterminate nodules as the research subject. The volume of the nodule is in the range of 50–500 mm3 that were re-examined at a 3-month follow-up CT examination. Based on a large amount of follow-up image data, we use recurrent neural network (RNN) to predict the changes of lung nodules. Because RNN can handle sequence input problems, the characteristics of nodules at baseline scans and follow-up scans as input of RNN. Therefore, we will use convolutional neural network (CNN) to extract features from the original CT image, and then input these features into RNN. 3 Future work According to the experimental design, our next step is to extract the features from the CT image after preprocessing, and the obtained features are fed into the RNN for training and testing. Finally, three traditional indicators (accuracy, precision and specificity), receiver operator characteristic and area under the curve are used to evaluate the proposed models.
Improvement of the axial velocity signal-to-noise ratio for shear wave elastography using a convolutional neural network
Xufei Chen, Nishith Chennakeshava, Ruud J.G. van Sloun, Massimo Mischi
Abstract: The Signal-to-Noise Ratio (SNR) of the axial displacement and its velocity can directly impact the accuracy of the Shear Wave Elastography (SWE) estimates, such as elasticity or viscosity. This SNR decreases when the push pulse pressure is lower, the focal depth is deeper, or due to attenuation when the shear wave is far from the push location. At the same time, minimizing the push pressure is of clinical relevance, as a lower mechanical index ensures safer clinical usage. Here, we exploit the denoising prior of a Convolutional Neural Network (CNN) that is stochastically trained to emulate the Loupas 2D autocorrelator, and obtain high SNR axial velocities (vz) from low push pressure in-phase and quadrature (IQ) data.
Denoising contrast-enhanced ultrasound sequences : A multilinear approach
Metin Calis, Alle-Jan van der Veen, Massimo Mischi, Borbála Hunyadi
Abstract: Contrast-enhanced ultrasound has shown promise for the localization of prostate cancer by detection of angiogenic changes in the microvascular architecture. The recent advances in three-dimensional contrast-enhanced ultrasound imaging enable the characterization of prostate microcirculation with a single intravenous injection of contrast agent. Although several markers of cancer angiogenesis have been extended to cover the three-dimensional ultrasound, most signal denoising algorithms do not exploit the inner structure of the volumetric data and instead vectorize the spatial dimensions, causing a loss of information about the location of the voxels. This work proposes a denoising algorithm based on the multilinear singular value decomposition and compares it to the singular value decomposition. The time-intensity curves are assumed to follow the local density random walk model, and the speckle noise is assumed to follow the Rayleigh distribution. The received log-compressed signal is modeled as the mixture of the time-intensity curves and the speckle noise, which can be approximated as white Gaussian noise with outliers. Multilinear singular value decomposition [1] is applied, the rank of the underlying signal is estimated using information-theoretic criteria [2], and the time-intensity curves are recovered using the components that define the signal subspace. The performance is analyzed using two simulation scenarios and an in-vivo study consisting of 6 patients who underwent prostatectomy. In the simulation scenario where only speckle noise is considered, the true multilinear rank of the signal is estimated correctly 87 out of 100 times. On the other hand, when various different noise sources such as motion artifacts are added to the signal model, the proposed algorithm fails to correctly estimate the ranks. However, the reconstruction error is still lower than applying singular value decomposition. The in-vivo study shows an improved voxel-based classification performance where nearly all the convective dispersion parameters have a higher area under the receiver operating characteristic curve. In summary, we have proposed a denoising algorithm for contrast-enhanced ultrasound sequences based on tensor decomposition. The proposed algorithm accounts for the multidimensional nature of the recordings and shows improved performance in classifying prostate cancer. The algorithm has the potential to be used in other despeckling applications.
Spatial summation investigated using intraepidermal electrical stimulation combined with adaptive psychophysical methods: A feasibility study
Niels Jansen, Boudewijn van den Berg, Jan Buitenweg
Abstract: Introduction. Spatial summation to nociceptive stimuli is a central mechanism which has been found to be altered in several chronic pain syndromes such as knee osteoarthritis and lateral epicondylalgia, while being unaltered in others such as fibromyalgia or low back pain. In this study, it is explored whether and how spatial summation could be evaluated via intra-epidermal electrical stimulation (IES) combined with adaptive psychophysical methods. Methods. 15 healthy subjects (11 female) with an average age of 22.8 (±3.7) were included in the study. Thresholds to three stimulus types were simultaneously tracked using adaptive psychophysical methods. Single-pulse (SP) stimuli were provided through either one of two IES-electrodes placed on the right volar forearm (~2 cm apart), or through both electrodes simultaneously. In order to adequately evaluate spatial summation with this protocol, the relative contribution of both the proximal and distal electrode must be approximately equal. Results. It was found that only in those participants where the contribution of both the proximal and distal electrode was (approximately) equal, spatial summation could be observed. In the participants where this assumption was not met, this was due to the threshold to the SP stimuli of one electrode becoming significantly higher than the other over the course of the experiment. Further, a positive relation was found between the stimulus intensity and the extent of spatial summation. Discussion. At the start of the experiment, unequal contributions to the SP stimuli to either one of the two IES electrodes were expected due to differences in the electrocutaneous interface. However, it was expected that these contributions would be unequal from the start – instead of over the course – of the experiment. These results provide relevant insights into the mechanisms resulting in habituation to the stimuli. Further based on the findings in this study, technical improvements are suggested and directions for future research are outlined.
Safe Neurostimulation Against Pain (SNAP)
Sofia Cecchini, Maurits Konings, Oda Heerema, Luuk Evers, Jesse Bosma, Albert Van Wijck
Abstract: Chronic neuropathic pain is cumbersome, difficult to treat and it induces a dramatic loss of Quality-of-Life in many patients. It is caused by a primary lesion or dysfunction of the nervous system and it is often treated using analgesic drugs and other conventional medical management that can lead to serious side-effects (dependence, drowsiness) and to which patients are not always responsive. Therefore, many efforts have been taken to find an alternative. Spinal Cord Stimulation (SCS) has been proven to be the most effective in suppressing chronic neuropathic pain: an electric stimulation current is applied in proximity of the vertical dorsal columns of the spinal cord which entails pain mitigation, according to the “gate control theory”. The current clinical SCS technique uses electrodes on a catheter that needs to be placed within the very vulnerable epidural space. This location is prone to infection, connective tissue formation and catheter migration. To overcome these drawbacks, a new concept (SNAP, Safe Neurostimulation Against Pain) has been formulated, in which the electrodes are placed on a clamp attached to the dorsal side of the thoracic vertebrae, and thus outside the potentially dangerous epidural space. The design comprises a new multi-electrode system on a small, stand-alone, wire-shaped implant that takes advantage of the electrically insulating property of the bone tissue of the spinal cord as a means of projecting the electrical stimulation current into the target area in the spinal cord. The first in-vitro simulations have shown promising results: the applied field appears to target the correct region in the spinal cord without substantial electric current leakages towards outside the foramen vertebrae. Thus, from these preliminary data it seems possible to achieve a sufficient pain mitigation effect using our safe and minimally invasive approach.
“DropAdjust” – A precise and accurate manually controlled over-line flow regulator for grafity infusion
Wouter Donders, Marit van Velzen, Edwoud Kooijman, Arjo Loeve
Abstract: INTRODUCTION Worldwide 70-90% of hospitalized patients receive intravenous infusion during their stay. A high degree of infusion accuracy is often essential, as deviation from the intended dose can be dangerous. Electronic infusion pumps provide the most accurate way of infusion, but they require programming and frequent maintenance. Furthermore, they are unsuitable for austere environments, are costly, and have shown to become scarce in times of a pandemic. Gravity infusion combined with a drop counter could pose an interesting alternative. However, this method appears to be inaccurate over time and setting an accurate flow rate is challenging. The typically used flow regulator, a roller clamp, is the cause of these drawbacks. This study aimed to solve these drawbacks by developing a precise and accurate manually controlled over-line flow regulator for gravity-driven infusion. METHODS & RESULTS The design process consisted of three design phases: analysis, synthesis and evaluation. During the analysis, the design requirements were set up and substantiated. The synthesis phase consisted of generating a morphological overview and conducting experiments for concept selections. Then, promising partial solutions were selected, and through rapid prototyping iterations a final design and prototype were created. In the evaluation phase, the flow regulator prototype performance was assessed based on the set requirements, including flow tests with rates occurring in clinical practice. The developed DropAdjust prototype demonstrated a major performance increase in terms of mean flow rate accuracy and regulation control compared to the roller clamp. CONCLUSION The DropAdjust satisfies all tested design criteria and outperformed the conventional roller clamp in terms of accuracy and precision. Moreover, it even enabled reaching a mean flow rate accuracy error comparable to infusion pumps. The prototype is ready for tests in practical situations, but several steps are still needed to realise a market-ready device. Endurance tests, usability tests and additional safety and functionality tests are next up. The DropAdjust, combined with a drop counter will provide a more affordable and accessible alternative to infusion pumps.
Specifications of haptic display for the knee
Praveen Kumar Pakkirisamy, Hans Timmerman, Elisabeth Wilhelm
Abstract: Total knee replacements (TKRs) are commonly performed to treat patients with osteoarthritis when pharmacological and conservative treatments do not provide adequate relief (Amanatullah, Rachala, Trousdale, & Sierra, 2014). In the UK, more than one hundred thousand knee replacement surgeries are done each year and a similar pattern is reported in many worldwide joint registries (Andrew J Price, 2018). Around 10-34% of patients with osteoarthritis are reported to face chronic postsurgical knee pain (CPKP) (Beswick, 2012). CPKP causes disability and is associated with loss of quality of life and increased use of resources for healthcare (Skrejborg, 2019). Robotic technologies are transforming rehabilitation from one-on-one human resource intensive treatment to a remotely supervised and low-cost enterprise (Laut, 2016). Most of these robotic devices focus only on motor training neglecting the sensory aspect of rehabilitation (Gassert, 2018). However, studies suggest that sensory retraining could facilitate recovery for the patients (Chia, 2019). For instance, a systematic review on sensory discrimination training for adults with chronic musculoskeletal pain concluded that sensory training has high potential regarding its clinical efficacy (Graham, 2020). In this research, we present a list of requirements to develop a haptic display for providing sensations such as touch, vibration, and heat to the knee. There exist specifications in the literature for the upper-limb haptic devices (Vidal-Verdú, 2007), however, no requirements exist for the development of knee-based haptic devices. This study can help in developing compact and ergonomic haptic displays for the knee which can be further used for sensory training purposes.
Armcoach4stroke project: The development and technical evaluation of a minimal sensing method for the measurement of arm exercise metrics in stroke patients
Ruben Regterschot, Peter Veltink, Bert-Jan van Beijnum
Abstract: Research question Wearable sensor-based methods are increasingly applied for the kinematic analysis of functional arm movements after stroke. As part of the ArmCoach4Stroke project, we are developing a minimal sensor-based method for the assessment of arm exercise metrics after stroke in the home environment. However, little evidence exists for the clinimetric properties of sensor-based arm exercise metrics (e.g., accuracy, test-retest reliability). The aims of our project are 1) to develop a minimal sensor-based method for the measurement of arm exercise metrics in stroke patients, and 2) to evaluate in healthy adults and stroke patients whether the accuracy and test-retest reliability of the sensor-based arm exercise metrics are sufficient. Methods The project consists of iterative steps of development, evaluation and optimization: 1. Focus group sessions with therapists to define important arm exercises and exercise metrics for stroke rehabilitation. 2. Define hypotheses for minimal sensor configuration to measure arm exercise metrics. 3. Collect data in several healthy adults to develop algorithms per minimal sensor configuration. 4. Evaluate per sensor configuration whether arm exercise metrics show sufficient accuracy and test-retest reliability in healthy adults in comparison to a Vicon reference system. 5. Optimize minimal sensor configuration and algorithms 6. Evaluate per sensor configuration whether arm exercise metrics show sufficient accuracy and test-retest reliability in stroke patients in comparison to a Vicon reference system. 7. Optimize minimal sensor configuration and algorithms. Preliminary results Focus group meetings with therapists showed that important arm exercises are: reach to point, reach to grasp, hold & release cylinder object, transport of cylinder object, push/pull/shove exercise. Focus group sessions with therapists revealed that important arm exercise metrics are: smoothness, velocity and movement trajectory of the end effector (hand), trunk flexion, shoulder elevation and abduction, range of motion, grip force, and object tilt during transport. The hypothesized sensor configuration consists of an IMU sensor on the hand or wrist, upper arm, and shoulder in combination with an instrumented cylinder object (IMU and grip force sensing). Development and evaluation of algorithms for the calculation of the metrics per arm exercise are in progress based on data of healthy adults.
Moving out of the lab: Can we predict vertical ground reaction forces with 3 intertial measurement units?
Bouke Scheltinga, Hazal Usta, Jasper Reenalda
Abstract: Running is a sport with a high injury incidence. Monitoring biomechanical load could help in understanding the development of these injuries. Ground reaction force (GRF) can be seen as an important measure to quantify biomechanical load during running [1]. However, GRF measurement is restricted to the lab. Artificial neural networks (ANNs) are capable to model complex relations and thus could be used to predict GRF from inertial measurement units (IMUs) [2]. This prediction would be a step towards the quantification of biomechanical load outside the lab. The goal of this abstract is to show the possibilities of a subject specific ANN to estimate GRFs in vertical direction using three IMUs. Methods: 6 experienced heel strike runners (3 F, 3 M; 31.5y ± 11.7y, 1.77m ± 0.07m, 60.7kg ± 16.0kg) ran 9 trials on a force-instrumented treadmill at combination of three velocities (10, 12 and 14km/h) and three stride rates (preferred, -10% of preferred and +10% of preferred). Subjects were instrumented with IMUs (240Hz) mounted at both proximal tibias and pelvis. With a subject specific approach, 40 strides were extracted per trial, 20% of this data was used as validation and 80% as training set. The trial at 12km/h at preferred stride rate was left out to test the model performance. Two layer ANNs (250 and 100 neurons) were then trained with the 3-dimensional gravity subtracted acceleration in the global frame as input to fit the vertical GRF. Performance of the models was analysed with the root mean squared error (RMSE) in body weight, Pearson’s correlation coefficient (r) and the mean of the absolute error of the vertical peak during stance as an percentage of the peak force. Results: The ANN modelled vertical peak GRF forces with high accuracy for 4 out of the 6 subjects (r>0.99, mean absolute peak error <2.5%, RMSE<0.094). Performance for the 5th and 6th subject was slightly worse (r>0.96, mean absolute peak error<4.2%, RMSE<0.23). Conclusion: Deployment of ANNs to predict vertical GRFs directly from gravity subtracted acceleration in the global frame is very promising with a subject specific approach.
Wearable to measure individual pitching biomechanics and its role in injury prevention
Bart van Trigt, Erik van der Graaff, Marco Hoozemans, Frans van der Helm, DirkJan Veeger
Abstract: Baseball pitching is an explosive and repetitive whole-body activity. It requires transfer of kinetic energy from the lower extremities through the trunk up to the upper extremities. The transfer of kinetic energy between body segments is described as the kinetic chain (Marshall & Elliott, 2000; Seroyer et al., 2010). The basics of the kinetic chain are that the energy derived from the rotation and translation of a segment can be transferred to the most distal segment. In pitching, the transfer of energy is optimal when segmental rotations are sequenced, which will – in comparison with incorrectly timed movements - result in higher ball speeds. To provide coaches and pitchers with insights about the pitchers’ segmental sequencing, there is a need to transfer and measure biomechanics from the lab into the field. We, therefore, developed the PitchPerfect system. This is a commercial based sensor system what measures the kinetic chain in baseball pitching. The PitchPerfect system, contains a short and shirt with removable sensors. The sensors measure the peak angular velocity of the pelvis (short) and trunk (shirt) and the timing of the peak angular velocity. The timing between both sensors can be used as an indicator to optimize the kinetic chain. The PitchPerfect system provide the pitcher with feedback in two ways: (1) Real time feedback of three biomechanical variables (peak angular velocities of pelvis and trunk and separation time) is presented directly on a mobile device after each pitch. (2) Detecting individual throws to calculate pitch count and workload over time. Throwing with high ball speeds increase the risk of sustaining an injury. Several studies investigated the relationship between biomechanical factors and injury risk (Anz et al. 2010; Hurd et al. 2011). However, none of these studies could investigate the cause-effect relationship because biomechanics were measured at a single moment in the lab. With the use of PitchPerfect it is possible to investigate the cause-effect relationship because it measures the individual pitchers’ biomechanics at multiple moments on the field. Furthermore, individual pitcher variability can be quantified, which might be important in relation to injuries (Trigt et al. 2020). Answers to these questions can be used as injury prevention and applied in an early warning system in the future.
Axially rigid active steerable needle for high-dose-rate prostate brachytherapy
Martijn de Vries, Jakub Sikorski, Sarthak Misra, John van den Dobbelsteen
Abstract: High-dose-rate (HDR) brachytherapy (BT) implant needles are rigid and restricted to linear insertion paths. Steerable instruments allow for precise access to deeply-seated targets while sparing sensitive tissues and avoiding anatomical structures. In addition, steerable instruments can enlarge the potential patient group eligible for HDR prostate BT, as generally patients with a prostate volume > 50-60 cm3 are excluded from this treatment modality due to pubic arch interference (PAI). In this study, we present and evaluate a novel omnidirectional steerable needle for HDR BT of the prostate. The instrument utilizes the commercial HDR BT outer catheter and an inner needle with internal compliant mechanism. This mechanism enables distal tip steering through proximal instrument bending while preserving high axial and flexural rigidity. Active steering of the instrument allows for adjustments of the catheter pathway and withdrawal of the inner needle creates a work channel for remote afterloading. Finite element analysis evaluates the design and the prototype is validated in experiments involving tissue simulants and ex-vivo bovine tissue. Ultrasound images are used to provide visualization and shape-reconstruction of the instrument during the insertions. Manually controlled active needle tip steering in inhomogeneous tissue simulants and ex-vivo tissue resulted in mean targeting errors of 1.4 mm and 2 mm in 3D position, respectively. We found lateral tip steering up to 20 mm. The experiments showed that the steering response of the instrument is history-independent. The results indicate that the endpoint variability of the steerable needle is similar to that of a conventional rigid HDR BT implant needle while adding the ability to steer along curved paths. High axial and flexural rigidity enable puncturing and path control within various heterogeneous tissues. The developed steerable instrument has the potential to overcome problems currently unavoidable with rigid HDR BT implant needles, such as PAI, without major changes to the clinical workflow.
Bedside visualisation tool for prediction of deviation from intended dosage in multi-infusion therapy
Robin Gevers, Maurits Konings, Agnes van den Hoogen, Annemoon Timmerman
Abstract: Background: In multi-infusion therapy multiple infusion pumps are connected to the same access point into the patient’s circulation. The different pumps in such a system influence each other, which leads to deviations from the set flow rate. It is generally known that these errors occur, however in practice, clinicians tend to find it hard to estimate an order of magnitude of these errors. Method: A panel of experts was gathered to obtain an indication of their level of knowledge about the deviations from the set flow rate in multi-infusion system. The panel, consisting of both nurses and doctors, was then shown the visualisation front-end of a model, that shows the pump settings and the deviations caused by changing the set flow rates. The model used is an analytical model containing the hardware parameters and the pump flow rate settings. Dosing deviations caused by the push-out effect (so called “dead volume”), a Poiseuille flow profile and mechanical compliance were individually calculated and combined. The system modelled consists of two infusion pumps one with liquid A and second with B, connected to a mixing point, followed by a catheter for vascular access to the patient. Results: A large fraction (44%) of our panel of experts wrongly predicted what effect changing the flow of liquid A has on the flow rate of liquid B that reaches the patient. The panel found it especially hard to estimate the deviation caused by the compliance of the material, as no one (0%) was able to correctly predict the change in flow rate of liquid B. This deviation is seen as counter-intuitive by the panel of experts. After the model was shown, the experts understood better what deviations to expect and are now able to apply this knowledge in clinical practice. Conclusion: Using a model to visualise the deviations from the set flow rate is a good opportunity to let clinicians obtain more knowledge about deviations from the set dosage in multi-infusion therapy. When using the model the deviations still occurred, however the clinician was now able to anticipate and take the deviation into account.
3D-printed porous cochlear implants
A. Isaakidou, I. Apachitei, L.E. Fratila - Apachitei, A.A. Zadpoor
Abstract: Drug administration is the cornerstone treatment for a variety of pathologies, with systemic administration being the most used method worldwide. However, drug bioavailability in the tissue of interest often does not reach therapeutic levels. Instead, the drug dissipates throughout the body contributing to undesirable secondary effects. Organs that possess a blood-barrier (e.g., cochlea) notably suffer from this phenomenon. Recently, researchers and pharmaceutical companies have proposed to mitigate this issue by using controlled, localized drug delivery systems where advances in the field of additive manufacturing enable their fabrication. In this study we used a high-resolution additive manufacturing method, namely two-photon polymerization (2pp), to fabricate a series of cochlear implant designs with internal microscale porosity at anatomically relevant sizes. We printed porous cochlear implants of 0.6x0.6x2.4 mm3 with two different pore sizes (i.e., 20 μm and 60 μm) and flat specimens of 2x2x0.5 mm3 for cell culture. We assessed the shape fidelity of these porous implants and macrophage viability on their constituent resin via scanning electron microscopy and fluorescent microscopy, respectively. The cell culture experiments showed no cytotoxicity of the 2pp resin for macrophages, validating the suitability of the method and the material for further drug incorporation and release, both in vitro and in vivo. The results of this study demonstrate that the 2pp process constitutes a feasible fabrication method for a new generation of cochlear implants with anatomically applicable sizes and appropriate porosity for drug loading. Keywords: Additive manufacturing, 2-photon polymerization, cochlear implant, porosity, cytotoxicity
State of the art cervical cancer brachytherapy: time action analysis and patient experience
Sharline van Vliet - Pérez, Rosemarijn van Paassen, Linda Wauben, Robin Straathof, Nick van de Berg, Jenny Dankelman, Ben Heijmen, Inger-Karine Kolkman-Deurloo, Remi Nout
Abstract: Objective Brachytherapy (BT) is an important component of the curative treatment for locally advanced cervical cancer. Patient experience in terms of pain, anxiety, and duration of each BT treatment step is still scarcely reported. The aim of this study is to perform a time-action analysis and determine the patient experience during each step of BT treatment of cervical cancer as benchmark and to understand and prioritise further improvements. Method In total 30 patients treated with 69 HDR BT fractions with an intracavitary/interstitial applicator were included for the time-action analysis of which 13 patients (28 fractions) were also included for the patient experience analysis. The time-action analysis included a standardised form with the reported time needed for each step. The patient experience analysis included an EQ-5D questionnaire with health state index (0= dead, 1= full health) and EQ VAS score (0= worst imaginable health, 1= best imaginable health) at the beginning of the day to establish a base line health status, and a numeric rating scale questionnaire (0= perfect situation, 10= worst possible situation) to assess the pain, anxiety and duration experience during each treatment step. The median and interquartile range for all parameters is reported. Results The total procedure time (hours:minutes) from arrival at BT department till discharge was 8:55 (8:00-9:25), which was used for: applicator implantation, recovery from operation, imaging, treatment delivery, applicator removal, and waiting time between each step. At the beginning of the day, patients had a health state index score of 0.82 (0.67-1.00) and EQ VAS score of 0.80 (0.63-0.88). Patients had the highest pain score between imaging and treatment delivery (3 (1-7)), the highest anxiety score during applicator removal (2.5 (0-8)), and the highest duration score between imaging and treatment delivery (6 (0-7.5)). The large variations in scores points at inter- and intra-patient variations. Conclusion This analysis highlights patient experience during different steps of cervical cancer BT workflow. In the future, the time-action and patient experience analysis can be used to optimise different steps of the BT treatment.
Image reconstruction via a model based neural architecture for intravascular ultrasound
Tristan Stevens, Nishith Chennakeshava,, Frederik de Bruijn, Andrew Hancock,, Martin Pekař, Yonina Eldar, Massimo Mischi, Ruud van Sloun
Abstract: Intravascular UltraSound (IVUS) probes are a major component in establishing the treatment and diagnosis of coronary heart diseases, such as abdominal aortic aneurysm, and artherosclerosis. Although it is widely employed, due to its highly cost-effective nature, it is still a highly challenging modality to interpret, with a steep learning curve. It suffers from a limited bandwidth, making a trade-off between spatial and temporal resolution inevitable. Users are also limited by the processing power available to them, to process the limited data that is obtained. Thus, we propose a model-based approach to the design of a neural architecture which aims to reconstruct a fully sampled image, from a channel under-sampled beamformed image. We utilise only 25% of the number of channels used by the fully sampled image. We achieve more consistent performance with our model when compared to a benchmark network with a similar number of free parameters. We achieve this by virtue of the neural architecture that was derived using a physical measurement model. As a result, we may now potentially reduce the volume data required to achieve a spatial resolution that is very comparable to the fully sampled images, in addition to retaining and enhancing the temporal resolution required for such tasks. Consequently, allowing us the ability to run multiple diagnostic modalities to enhance the usability of IVUS to the clinician.
Decreasing operator-dependency in carotid ultrasound measurements: clinical study plan
Esmee de Boer, Luuk van Knippenberg, Catarina Dinis Fernandes, Rohan Joshi, Sabina Manzari, Arthur Bouwman, Massimo Mischi
Abstract: Hemodynamic monitoring is of utmost importance when treating critically ill patients. Currently used hemodynamic techniques are invasive, such as the transpulmonary thermodilution approach for cardiac output estimation, and their use brings the risk of catheter-related complications. Over the last two decades, carotid artery ultrasound (US) has been investigated as a non-invasive modality for hemodynamic monitoring, including the estimation of cardiac output. Traditionally, carotid flow measurements are performed with the US probe oriented parallel to the vessel. Assuming a circular cross-section of the vessel and a parabolic flow profile, the probe should be properly positioned along the mid-axis to obtain an accurate blood flow calculation. However, it is difficult to obtain and maintain this midaxis parallel view, and the literature describes that operator experience may impact the reliability of carotid flow measurements. Another way of assessing the cross-section of the carotid artery is by rotating and tilting the probe, a view that is easier to visualize and assess for sonographers. As such, we hypothesize that a different, i.e., rotated and tilted, orientation of the probe results in a decrease in operator-dependency while preserving accurate diameter and velocity estimates. Preliminary research pointed out that it is feasible to perform velocity estimations using a rotated US probe orientation, and showed that this view is more robust to motion and less operator-dependent compared to parallel measurements. To test this hypothesis, data of 25 adult cardiac surgery patients will be obtained. Pre-surgery, we will perform carotid Doppler ultrasound measurements with the probe oriented parallel, transverse, and rotated and tilted. As flow is calculated from diameter and velocity estimates, we will compare these estimations obtained from the different probe orientations to ascertain equivalence in a within-patient comparison. As data collection is ongoing, no results can be shown yet.
A study of the sensitivity of carotid haemodynamic indices to cardiovascular variation in a virtual population
Irene Suriani, R. Arthur Bouwman, Massimo Mischi, Kevin D. Lau
Abstract: The analysis of carotid ultrasound (US) flow, velocity, and diameter waveforms provides important information about cardiovascular and circulatory health. These waveforms can be used to derive clinical indices of atherosclerosis, vascular aging, and, potentially, haemodynamic state (e.g, cardiac output, fluid responsiveness). The morphology of carotid US waveforms is the result of a complex interaction between central haemodynamic variables (e.g., heart rate, stroke volume, left ventricular ejection time), and arterial properties (vessel geometry and stiffness, peripheral vascular resistance and compliance, and wave reflections from peripheral body sites) [1]. The use of one-dimensional (1D) haemodynamic modelling is a computationally inexpensive approach that provides a tool to investigate how waveform morphology and derived clinical indices relate to local and global arterial, and cardiac properties. Based on in-vivo measurements of cardiac parameters and arterial diameters, lengths and stiffnesses, 1D models of the arterial tree can reproduce realistic pressure, velocity, diameter and flow waveforms throughout the arterial tree [2]. In previous work [3], we clinically validated a virtual population (VP) that realistically simulates carotid waveforms of healthy subjects aged between 10 and 80 years old, obtained by varying arterial and cardiac parameters of a 55-branch model [2] described using literature values for age-related variations. In particular, the definition of carotid-specific diameter and stiffness variations were improved compared to previous VPs[4, 5]. In the present work, we extended this VP, introducing realistic intra-subject variations per age group by simulating the documented standard deviations of parameters such as diameter, stiffness, etc. The sensitivity of clinically relevant carotid indices (e.g., pulsatility index, resistance index, flow augmentation index) in response to these variations was then analysed. Furthermore, the VP was used to investigate how wave reflections from different arterial bifurcations affect the morphology of carotid waveforms and derived indices. This work can help clarify the clinical interpretation of carotid waveform indices by inferring the physiological mechanisms underlying the morphology of measured carotid US waveform. [1] J. P. Mynard, et al., “Measurement, Analysis and Interpretation of Pressure/Flow Waves in Blood Vessels”, 2020. [2] J. Alastruey, et al. , “Arterial pulse wave haemodynamics”, 2012. [3] I. Suriani, et al., “Validation of an aging virtual population for the study of carotid hemodynamics”, 2021 (IEEE conference proceedings, accepted for publication). [4] M. Willemet, et al., “A database of virtual healthy subjects to assess the accuracy of foot-to-foot pulse wave velocities for estimation of aortic stiffness,”, 2015. [5] P. H. Charlton, et al., “Modeling arterial pulse waves in healthy aging: a database for in silico evaluation of hemodynamics and pulse wave indexes”, 2019
Deep reinforcement learning on cerebral anterior tree vessel tracking
Jiahang Su, Shuai Li, Lennard Wolff, Geert Lycklama à Nijeholt, Wim van Zwam, Wiro Niessen, Aad van der Lugt, Theo van Walsum
Abstract: Background: Cerebral anterior vessel trees are relevant for understanding ischemic stroke and many chronic diseases. Quantification of vessel tree parameters requires the extraction of the anterior vessel trees, e.g. from CTA images. We investigate to what extend a Deep Reinforcement Learning (DRL) based approach can be used for cerebral vessel tracking in CTA images. Method: The DRL tracker was formulated as directed tracking, starting from an initial position and direction. The state of the tracker in our case was the concatenation of three consecutive 21x21x21 sub-volumes centered along the tracking path. The action space was defined as the 26 connected neighbouring voxels of the current position with a scaler factor of 2. The instant reward was defined using a surface distance-based metric and was further enhanced using a Dice overlap metric. The optimal policy was obtained using a policy gradient based proximal policy optimization approach with advantage actor-critic framework. The network was configured using a classical dilated network. The proposed DRL vessel tracker was trained with 7000 vessel centrelines from 80 subjects. We further use an independent validation set of 10 subjects and a test set of 20 subjects. The results were compared with a baseline Deep Q Network (DQN) vessel tracker. Results: The DRL vessel tracker was evaluated using 5 nested fold cross validation. Those 5 nested folds uses same validation sets. The average curve to curve similarity metric achieves 0.7 for 100 subjects. The average overlap till the first error reaches 0.9 on 100 subjects, where an error was defined as a distance larger than 1.6mm off the centreline path. The tracking success rate was 0.7. The baseline DQN tracker has curve to curve distance metric of 0.4. Conclusion: A DRL vessel tracker can be used to detect cerebral anterior arteries in CTA images of the brain.
Phantom-based coronary flow measurement during systole with ultrafast Doppler imaging and diverging waves
Yizhou Huang, Xufei Chen, Emilia Badescu, Ruud van Sloun, Massimo Mischi
Abstract: The advent of ultrafast Doppler imaging has led to the redefinition of coronary ultrasound capabilities, enabling high Doppler sensitivity and more accurate flow measurement compared to conventional Doppler. Recent studies [1] showed the great potential of ultrafast coronary imaging using linear probes and plane-wave imaging combined with singular value decomposition (SVD) for clutter filtering during the diastole phase. Here, we aim at verifying the feasibility of using a phased array and diverging wave imaging to measure coronary flow in a phantom-based environment with fast clutter motion. The performance of SVD with different clutter-blood thresholding methods were compared. A 15% gelatine phantom with a 3-mm diameter channel at 8 cm depth was made to mimic the coronary flow measurements. A pulsatile pump was used to generate flow at 5, 10 and 40 cm/s speed. One-second ultrasound data was acquired with a Verasonics Vantage system and a Philips S5-1 phased array probe. A motion generator was connected to the probe to generate a sinusoidal axial motion at 6 cm/s maximum. A pulse repetition frequency of 4400 Hz with a centre frequency of 3.125MHz, and an 8-angle alternate sequence (-7º to 7º with a step of 2º) was used. SVD with three thresholding methods introduced by Maresca et al., [1] and Baranger et al., [2], as well as a novel contribution proposed in this abstract for adaptive determination of the turning point of the singular value curve, were investigated. Contrast-to-Noise Ratio (CNR) was used as a quantitative performance metric. SVD with our proposed thresholding method provided a 10.2 dB CNR when the flow was at 40 cm/s, mimicking measuring coronary flow from a healthy adult during the early diastole phase. Moreover 5.7 dB CNR was still preserved when the flow (5 cm/s) was close to the generated motion, mimicking the situation during the systolic phase. Future investigation will focus on the performance in vivo.
Personalized risk assessment of carotid artery stenosis by ultrafast ultrasound flow imaging
Janna Ruisch, Suzanne Holewijn, Michel Reijnen, Chris de Korte, Anne Saris
Abstract: Approximately 20% of strokes originate from a plaque rupture in the carotid artery. Patient selection for surgical plaque removal, carotid endarterectomy, is far from perfect; the number-needed-to-treat in patients with severe stenosis is six to prevent one stroke within 5-years. Nowadays, ultrasound is the technique of choice for treatment planning, which is based on simple geometrical and blood velocity measures. However, conventional velocity measurements show large variability and low frame rates restrict complex or high blood velocity measurements. Especially these complex flow patterns, and the resulting wall shear stresses (WSS) acting on the vessel wall, are crucial throughout all stages of atherosclerosis. The advent of ultrafast ultrasound makes continuous tracking of flow in all directions feasible. The high-frame-rate acquisitions opened up new possibilities for 2D blood flow imaging (BFI) in the carotid artery, using cross-correlation-based speckle tracking. In this project, we aim to assess if and which blood flow velocity (related) parameters, such as WSS, vorticity and vector complexity, are associated with plaque progression and vulnerability using non-invasive, ultrasound-based BFI. Multiple explorative clinical studies will be performed to unravel the relation between blood flow velocity profiles and the progression and risk of carotid plaques. First, a validation study will be performed to evaluate the performance of BFI compared to 4D flow MRI in twenty healthy volunteers. Besides, the inter- and intra-observer variability of this novel BFI technique will be determined. Second, BFI will be performed in asymptomatic patients (N=85) with 30%-70% carotid stenosis during 2-year follow-up. Two subgroups will be established based on conventional ultrasound measurements: stable plaque versus plaque progression. The association between blood flow velocity (related) parameters and disease progression will be evaluated. Lastly, BFI will be acquired in patients (N=70) with >50% carotid stenosis who are scheduled for carotid endarterectomy. Histology staining of the excised plaque will differentiate vulnerable from stable plaques. The association between blood flow velocity (related) parameters and plaque vulnerability will be examined. With these explorative studies, we aim to demonstrate the potential of ultrasound-based BFI in the assessment of plaque progression and vulnerability. In future, this might enable more sophisticated, personalized treatment decision-making.
ultrasound carotid artery flow simulation using the k-wave toolbox
Yuyang Hu, Michael Brown, Didem Dogan, Geert Leus, Pieter Kruizinga, Johannes G. Bosch
Abstract: Background and Aim: We are developing a compressive sensing ultrasound device using spatial coding masks for carotid artery (CA) flow monitoring. For the required simulations, k-Wave is adopted because of its capability to simulate wave propagation in arbitrary distributions of heterogeneous material properties, including non-linear effects and attenuation. However, this toolbox is not designed for modelling the motion of medium or flow. In this study, we realize a time-varying k-Wave model of the CA and verify the validity of the resulting B-mode images and color flow Doppler (CFD) estimation. Methods: In our CA simulation model, a 6 mm wide vessel tilted at 15° is located within a 40×30 mm^2 tissue region. Inside the vessel, blood is flowing with a parabolic velocity profile and a constant peak velocity of 1 m/s. When modelling in k-Wave, the medium is described as a discrete grid with speed of sound and density properties at each grid point (rather than a distribution of scatterers). For our CA configuration we generate a sub-wavelength (0.05 mm) grid that contains random samples from a normal distribution for speed of sound and density corresponding to real tissue (1547 m/s,1050kg/m^3,σ=1.5) and blood (1584 m/s,1060kg/m^3,σ=0.1). The resulting random pattern for t=0 is displaced for the following frames using the velocity field and a new discrete grid distribution is calculated by interpolation and down-sampling to the k-Wave grid. Results: The simulation was evaluated by reconstructing B-mode and CFD images from the received signals over 30 frames by delay and sum (DAS) and autocorrelation, respectively. The resulting B-mode image shows intensity distributions for tissue and blood as expected (CNR = 1.46). The resulting CFD shows the velocity profile as expected, where the peak velocity (1.08 m/s) is slightly overestimating the simulation (1 m/s) and the RMS difference from the true parabolic velocity profile over the cross-section of the vessel is 0.14 m /s. Conclusion: we proposed a CA flow simulation with the k-Wave toolbox. The reconstruction of received signals shows that this model provides a realistic B-mode and CFD behavior. The measured parabolic velocity pattern corresponds well to the simulated flow. This model will be a useful tool for our future study of compressive ultrasonic sensing of flow in the carotid artery.
Ultrasound blood velocity imaging in the carotid artery using cascaded dual-polarity waves
Joosje de Bakker, Stein Fekkes, Chris de Korte, Anne Saris
Abstract: Stroke is a leading cause of death and major disability. Atherosclerosis is one of the primary causes of stroke. Complex blood flows in the carotid arteries seem to play a crucial role in the atherosclerotic disease process. However, clinically used ultrasound systems cannot measure these complex flows. The introduction of ultrafast ultrasound, where thousands of images are acquired per second, enabled new opportunities for blood velocity imaging. Identification of complex carotid flows could aid in better patient-specific diagnosis, risk stratification, and treatment planning. Ultrasound blood velocity imaging benefits from using higher transmit frequencies due to the increased spatial resolution, however, at the cost of penetration depth. Cascaded Dual-polarity Waves (CDW)a can compensate for the latter by using a train of pulses instead of a single pulse. However, its application in blood velocity estimation has not been shown yet. The performance of 8-CDW was evaluated and compared to single-pulse Plane Waves (PW) using Field IIb,c simulations and experiments of laminar flow. Different attenuation situations were mimicked by adding different noise levels to the simulations and by using a PDMS slab (14.4 dB) during the experiments. Data were acquired with a 7.8 MHz linear array transducer connected to a Verasonics Vantage 256 research ultrasound system. Steering angles of repeatedly -20° and +20° were transmitted at a repetition frequency of 8 kHz. A normalized two-step cross-correlation based compound speckle tracking method was used to determine the blood velocityd. Both simulations and experiments showed that CDW outperforms PW velocity estimations under more demanding attenuating situations. Without attenuation, the experiments revealed a mean relative velocity bias of 6.3% and 9.3% for PW and CDW respectively. This bias increased to 71.6% and 9.5% when adding the attenuating slab. Here, CDW outperformed PW due to the increased signal-to-noise ratio (SNR). However, since blood motion in between different pulses of CDW results in imperfect decoding, i.e. the imperfect summation/cancellation of shifted pulse responses originating from the pulse-trains, CDW does not outperform PW in case of sufficient SNR. To conclude, CDW could potentially aid in velocity imaging in high attenuating situations, such as in obese patients. REFERENCES aZhang, Y., Y. Guo, and W.N. Lee, Ultrafast Ultrasound Imaging With Cascaded Dual-Polarity Waves. IEEE Trans Med Imaging, 2018. 37(4): p. 906-917. bJensen, J., FIELD: A program for simulating ultrasound systems. Medical and Biological Engineering and Computing, 1996. 34: p. 351-352. cJensen, J.A. and N.B. Svendsen, Calculation of pressure fields from arbitrarily shaped, apodized, and excited ultrasound transducers. IEEE Trans Ultrason Ferroelectr Freq Control, 1992. 39(2): p. 262-7. dSaris, A., et al., In Vivo Blood Velocity Vector Imaging Using Adaptive Velocity Compounding in the Carotid Artery Bifurcation. Ultrasound Med Biol, 2019. 45(7): p. 1691-1707.
Ultrasound-based Fluid-Structure Interaction modeling of Abdominal Aortic Aneurysms: model verification and personalization
Judith Fonken, Floris Verheijen, Hein de Hoop, Esther Maas, Arjet Nievergeld, Marc van Sambeek, Frans van de Vosse, Richard Lopata
Abstract: Research on abdominal aortic aneurysm (AAA) development, growth and rupture risk requires a longitudinal study on biomechanical analysis of the AAA in a large set of patients. Since multiple studies have demonstrated that both wall mechanics and hemodynamics are highly dependent on AAA geometry, a patient-specific risk assessment is required, based on fluid-structure interaction (FSI) models, since the deformation of the AAA tissue is influenced by the hemodynamics in the AAA, and vice versa [1]. Time-resolved 3-dimensional ultrasound (3D+t US) is the preferred image modality to extract the patient-specific geometry, since it is safe, fast and affordable. A previous study has shown the feasibility of 3D+t US-based FSI simulations and the importance of incorporating the pre-stress [2]. Experimental verification of the FSI modeling framework is performed. Using a mock circulation loop [3], a blood-mimicking fluid and an elastic, 3D-printed, patient-specific phantom, the in-vivo hemodynamics are mimicked and matched to the prescribed FSI boundary conditions. During the experiment, high-framerate US acquisitions are obtained. Employing speckle-tracking methods, the wall displacement and velocity field are obtained and compared to the results of the FSI simulation, showing good correspondence. Due to the limited field-of-view of 3D+t US, the aorto-iliac bifurcation is often not included in the 3D+t US acquisitions. It has been shown that the presence of the aorto-iliac bifurcation does not significantly influence the wall mechanics in the AAA region [4]. However, previous studies have shown that the bifurcation does influence the hemodynamics in the AAA. Part of our study aims at quantifying the influence of excluding the bifurcation in FSI simulations. Here, patient-specific geometries are derived from CT-scans of 20 patients, to include the aorto-iliac bifurcation. For each patient, an additional simulation is executed, in which the bifurcation is replaced by a single outlet. The results in the aneurysm region are compared to quantify the influence of the bifurcation. In future studies, the FSI framework will be personalized by including the patient-specific shear modulus [5] and inlet velocity profile. The envisioned framework for personalized 3D+t US-based FSI simulations paves the way for future longitudinal studies on AAA development, growth, and rupture risk.
A phenomenological model for blood pressure prediction from NIRS signals
Theodore Nikoletopoulos, Marjolein Klop, Anthony C.C. Coolen, Richard J.A. van Wezel
Abstract: Orthostatic Hypotension (OH) is a common condition among older adults, characterized by a prolonged and/or deep blood pressure (BP) drop upon standing up. Standing up causes gravity-induced venous pooling in the legs and as a response BP drops. Feedback mechanisms in the body force BP to recover, however slow recovery may lead to symptoms such as dizziness and light headedness which may cause falls. Thus the rate and depth of decline in BP after standing up and the time to recovery are of interest, and to this end, methods that enable continuous BP monitoring are required. A suitable technique might be Near Infrared Spectroscopy (NIRS) which can detect changes in oxygenated and deoxygenated haemoglobin in the brain. The resulting continuous time signals are considered good markers of OH occurrence. In this work, we present a phenomenological model which describes reasonably accurately the dynamics of BP and NIRS signals, as well as the bio-mechanical adaptation effects that lead to BP recovery, taking into account postural changes. We use a probabilistic formulation and calculate the joint likelihood of observing BP and NIRS signals and the conditional probability of future BP values given past NIRS values. The latter object enables us to provide forecasts of blood pressure evolution with error bars and estimates of rate and depth of BP drop and time to recovery. Furthermore, most of the model’s parameters are physically interpretable, since they describe charcteristic relaxation times of the dynamics and could hence prove to be of clinical interest. We performed extensive numerical experiments with synthetic data, which indicate that the parameter inference and prediction algorithms based on our model are robust. Current testing of our model on real data is ongoing. This work is part of the EU funded project PROHEALTH (OP-Oost) which aims at detecting and preventing OH by wearable NIRS sensors, both in a laboratory and in a home setting.
Investigating the impact of collagen orientation on the local mechanical behaviour of atherosclerotic plaque caps
Hanneke Crielaard, Tamar Wissing, Ranmadusha Hengst, Ali Akyildiz, Frank Gijsen, Gert-Jan Kremers, Carlijn Bouten, Kim van der Heiden
Abstract: Stroke is commonly initiated by rupture of the atherosclerotic plaque fibrous cap in a carotid artery. However, cap rupture mechanisms are not well understood yet. Collagen is the main load-bearing component in plaque caps. Understanding its impact on local cap mechanics may provide critical insights into plaque rupture. Various limitations within in vivo and ex vivo animal and human plaque studies highlight the need for additional methods to investigate rupture mechanics. Therefore, we created tissue engineered collagenous plaque cap analogs, with controllable collagen architecture composition, that can be imaged and mechanically tested [1]. In the current study we present our pipeline for visualizing collagen orientation and consecutively obtaining local mechanical properties in these analogs to analyse the relation between collagen architecture and cap properties. Five cap analogs were created [1] and subsequently incubated with an Oregon Green labelled CNA-35 probe, which binds to collagen including newly formed collagen fibrils [2]. Multiphoton microscopy (MPM) to image CNA-35 and MPM with second harmonic generation (SHG) were performed simultaneously to visualize the collagen architecture. After imaging, the analogs were exposed to uniaxial tensile tests until failure. Next to conventional gauge length-based strains, digital image correlation (DIC) was performed to obtain local strains. Combining both imaging methods provided a more complete overview of the collagen architecture than each method independently. With SHG, larger imaging depth was reached. CNA-35 visualized additional newly formed fibrils in the surface layer. DIC analysis demonstrated that the gauge length-based strain measurements underestimate the rupture strain. DIC-driven local measurement showed that regions of elevated tensile/shear strain coincide with the rupture locations. We are currently analysing the correlation between collagen orientation and local strains. [1] Wissing TB, Van der Heiden K, Serra SM, Smits AIPM, Bouten CV, Gijsen FJH . Tissue-engineered collagenous fibrous cap models to systematically elucidate atherosclerotic plaque rupture. BioRxiv. Preprint. 2021 doi:10.1101/2021.07.20.451997; [2] Krahn KN, Bouten CV, van Tuijl S, van Zandvoort MA, Merkx M. Fluorescently labeled collagen binding proteins allow specific visualization of collagen in tissues and live cell culture. Anal Biochem. 2006 Mar 15;350(2):177-85. doi: 10.1016/j.ab.2006.01.013.
Comparison of arterial input functions derived from dynamic contrast enhanced MRI and Dynamic susceptibility contrast MRI
Chih-Hsien Tseng, Jaap Jaspers, Alejandra Mendez Romero, Thijs van Osch, Frans Vos
Abstract: Introduction The Arterial Input Function (AIF) plays a crucial role in estimating perfusion properties from Dynamic Susceptibility Contrast MRI data (DSC-MRI). An important issue, however, is measuring the AIF’s absolute contrast agent concentration due to uncertainties in the relation with the measured R2*-weighted signal. A potential solution could be to derive the AIF from separately acquired Dynamic Contrast Enhanced MRI (DCE-MRI) data. DCE-MRI facilitates absolute quantification of the contrast agent concentration through an established model. In this work, we aim to (1) visually compare the AIF determined from DCE-MRI with the AIF from DSC MRI in back-to-back acquisitions; and (2) asses the linearity of the relation between the transverse relaxivity change and the contrast agent concentration. Materials and methods Data acquisition: MRI datasets from fourteen patients were prospectively acquired both before and after radiotherapy as part of the associated RIGEL study (Trial identifier: NCT04304300). DCE imaging immediately succeeded by DSC imaging, were performed on a 3T MRI scanner (Signa Premier, GE) in Erasmus Medical Center. Data analysis: A few Regions-Of-Interests (ROIs) were manually selected in several arteries across several planes in the DCE images (these selected ROIs were then projected to the registered DSC images). Subsequently, the arterial signals from DCE and DSC were plotted to investigate the correlation between the signals. To investigate the linearity of MR signal change and contrast concentration change, we compared the R2* changes in the DSC images with the contrast concentration change derived from DCE images by vessel-wise comparison. Results We have found that the DCE-derived AIF significantly correlates with DSC-derived AIF, and yields more stable measurements than the DSC-derived AIF among different voxels in the same artery (less variation in peak height while sharing similar patterns). In addition, the R2* change was significantly correlated (R-squared value > 0.8) to the contrast concentration. However, this linearity was not significant while the relation was assessed in individual voxels, possibly due to noise and complex signal evolution in the DSC measurements. Conclusion The DCE-driven AIF correlates significantly with the AIF form DSC MRI. This might indicate that the quantitative DCE AIF could replace the DSC-driven AIF.
Determining abdominal aortic aneurysm rupture risk-predicting parameters from 4D ultrasound: method development and application
Esther Maas, Joerik de Ruijter, Judith Fonken, Arjet Nievergeld, Mirunalini Thirugnanasambandam, Marc van Sambeek, Richard Lopata
Abstract: In current clinical practice, the rupture risk of an abdominal aortic aneurysm (AAA) is estimated from its maximal diameter. Previous studies have however shown that additional parameters, such as AAA volume, curvature, maximal wall stress, and elasticity [1-4] could improve risk stratification by enabling personalized rupture risk estimation. Earlier research has mainly been based on computed tomography (CT) data [1-3], which has the disadvantage of radiation and use of nephrotoxic contrast, preventing follow-up studies. The non-invasive character and low cost of 4D ultrasound (US) do enable longitudinal studies, which makes it the preferred image modality. The aim of this study is to 1) develop a method to automatically determine all these rupture-risk related parameters from 4D US acquisitions and 2) apply this in a follow-up study of AAA patients, relating these parameters to AAA growth and rupture risk. A workflow has been developed to automatically segment the aortic inner wall from 4D US acquisitions, using an adaptation of a Star-Kalman algorithm [5]. The resulting geometries show good correspondence to CT geometries (golden standard), with a median Hausdorff distance of 5 [4.4-5.6] mm (median [IQR]) and similarity index of 0.89 [0.86-0.92] (n= 34). From these geometries, the curvature, maximal diameter perpendicular to the centerline, and AAA volume are determined. The segmentation is repeated on all time frames of the 4D ultrasound data. The displacement obtained from these segmentations, combined with the patient-specific blood pressure, is used to determine the pressure-strain elastic modulus. The geometry also serves as an input for finite element modelling (FEM), where the patient-specific blood pressure is applied to the inner vessel wall. This model provides aortic wall strains and 99th percentile stresses. Furthermore, the patient-specific wall stiffness is determined with an iterative updating FEM approach, combined with displacements from the US data. This fully automatic workflow is now being applied to a subset of the 500 patients in our clinical follow-up study. With this, relations between the beforementioned parameters and the AAA growth and rupture risk are studied. To our knowledge, this is the first study investigating all these relevant parameters concurrently in a follow-up patient study. [1] Lindquist Liljeqvist et al. 2016, J. Vasc. Surg [2] Lee et al. 2012, Ann Biomed Eng [3] Fillinger et al. 2002, J Vasc. Surg [4] Wilson et al. 2003, J. Vasc. Surg [5] Guerrero et al. 2007, IEEE Trans Med Imaging
Multi-perspective 3D ultrasound imaging of the aorta using sparse matrix arrays
Hein de Hoop, Marieke Vermeulen, Hans-Martin Schwab, Richard Lopata
Abstract: In the last decade, developments in 3D ultrasound imaging have allowed us to capture the partial 3D geometry and deformation of abdominal aortic aneurysms, which can aid clinicians in rupture risk assessment through patient-specific mechanical information of the vascular wall. The estimation of these parameters is mostly global, as the image quality at the sides of the aortic wall is degraded by the low lateral resolution of ultrasound and lack of contrast. Expanding the imaging aperture with an additional probe brings opportunity to improve the spatial resolution and extend high quality information over a larger part of the vessel's circumference. This study investigates the feasibility and merit of this technique ex-vivo in a realistic model of the abdomen. Two synchronized research ultrasound systems were each equipped with a 32×32 element matrix array (3 MHz, Vermon). An interleaved transmission scheme was designed where each transducer subsequently transmits 25 diverging waves steered around the z-axis, using sparse random apertures for high volume rates and increased heterogeneity of the acoustic field between transmits. Each transmission was received by both transducers and reconstructed into volumetric RF data (90 Hz/transducer) after the acquisition and registration. The image resolution was assessed with a 3D autocorrelation of speckle in a gelatin phantom containing scatter and position markers. The contrast-to-noise ratio (CNR) was quantified on the vascular wall of a dynamic porcine aorta phantom including anatomical features. The full-width-half-maximum (FWHM) of the 3D autocorrelation volume, i.e. speckle size, in compounded multi-perspective images was reduced by 66 % compared to the single perspective images at a depth of 8 cm. The long axis eccentricity of the FWHM decreased by 8%, showing a more isotropic resolution. In the aorta phantom, angular coverage of the vascular wall reflection was vastly increased and the CNR between wall and lumen increased by 4.5 dB compared to the single perspective image despite the accumulation of noise in the lumen. Future work will focus on the assessment of 3D deformation patterns, generic image registration methods, and the translation to in vivo.
Ultrasound-based automatic modulography for atherosclerotic artery plaques
Koen Franse, Jan-Willem Muller, Marc van Sambeek, Richard Lopata
Abstract: Introduction Currently the risk of carotid atherosclerotic plaque rupture is assessed solely based on the diameter reduction of the artery. However, research has shown that many rupture events are not correctly predicted by this criterium. Patient-specific biomechanical modelling can give more insight in the cause of rupture and act as a clinical decision support tool. The goal of this study is to develop an automatic inverse finite element (FE) modulography workflow that can reconstruct the heterogeneous characterization of the atherosclerotic artery wall based on non-invasive ultrasound (US) scans, creating patient-specific input for fluid-structure-interaction (FSI) modelling. Methods The input for the inverse FE workflow is 2D US strain imaging data of the arterial plaque cross-section. From this geometry, a homogeneous FE model is created for which the stress is computed. The ratio between the FE stress and the US strain is used to compute an apparent Young’s modulus (AYM) locally, which is further optimized to minimize the error between the US strains and the iteratively computed FE strains [1]. Subsequently a framework was developed for realistic US simulations of carotid plaques, based on 2D FE models and k-Wave simulation software. Strain imaging can be performed on these US simulations, to create more realistic strain-input data for the inverse FE workflow while being able to compare it to the ground-truth material properties and strains from the FE model. Results & Discussion Testing the inverse FE workflow by using FE strain input verified correct implementation compared to previous research [1] and showed correct segmentation and stiffness estimation in in silico carotid plaque data. The inverse FE workflow tested on FE strain input data showed promising results for local material parameter estimation in arterial plaques. However, further improvement is needed for correct performance on more realistic US strain data. Once this is achieved in 2D, the inverse FE workflow will be tested on 3D data. The final goal is to use the material parameters resulting from the inverse workflow as an input for patient-specific modelling of atherosclerotic carotid arteries. [1]: J. Porée, B. Chayer, G. Soulez, J. Ohayon and G. Cloutier, “Noninvasive Vascular Modulography Method for Imaging the Local Elasticity of Atherosclerotic Plaques: Simulation and In Vitro Vessel Phantom Study”, IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control, Vol. 64, No. 12, pp. 1805-1817 December 2017
Prediction of analog thrombi composition and mechanical properties using CT imaging
Jo-Anne Giezen, Janneke Cruts, Frank Gijsen, Kim Van Gaalen, Robert Beurskens, Yanto Ridwan, Marcel Dijkshoorn, Heleen Van Beusekom, Nikki Boodt, Aad Van der Lugt, Rachel Cahalane
Abstract: Background Stroke is caused by a thrombus that blocks an intracranial artery. The efficiency of a thrombus removal procedure is thought to be significantly influenced by the thrombus mechanical properties [1]. If the thrombus mechanical behaviour can be predicted from pre-interventional radiological imaging characteristics, the most suitable patient-specific treatment can be selected. Therefore, we will investigate the direct relationship between computed tomography (CT) imaging characteristics and thrombus mechanics. Methods Thrombus analogs were made from drawn blood of three healthy donors, and twenty clots per donor were made with five different volumetric red blood cell (RBC) ratios: 0%, 20%, 40%, 60% and 80%. The resulting composition of the thrombi was confirmed by histology. Clinical CT imaging was performed to measure the thrombus density. Perviousness, which reflects the permeability of a thrombus, was quantified by measuring the thrombus CT density after the administration of a contrast agent. The compressive stiffness of the thrombi at 75% strain (E75%) was obtained by performing unconfined compression tests. Results Histological analysis showed that the initial volumetric RBC ratios of 0%, 20%, 40%, 60% and 80% produced thrombi with non-equivalent compositions of 0.0%, 89.8%, 95.1%, 97.4%, and 99.4% RBC, respectively. This indicates that only 0% RBC and RBC-rich thrombi (>80% RBC) were produced. CT imaging showed that 0% RBC thrombi had a mean density of 36 HU and a density increase of 88 HU twenty minutes after the contrast agent was administered. The RBC-rich thrombi had a higher mean density of >81 HU, and were less pervious with a density increase of <14 HU. The mechanical tests showed that E75% decreased with increasing RBC content: e.g. for one donor we measured E75% of 44.4, 15.4, 12.6, 6.2, and 3.8 kPa for increasing RBC content. We also observed a large variation between donors for thrombi with comparable composition. For instance, for the 40% RBC thrombi we quantified E75% of 12.6, 72.4, and 5.6 kPa for the three donors. Conclusion We can conclude that thrombus composition affects both CT imaging characteristics and mechanical properties.
Suppressing tissue clutter in the presence of motion and nonlinear propagation of ultrasound
Geraldi Wahyulaksana, Luxi Wei, Antonius F.W. van der Steen, Hendrik J. Vos
Abstract: Introduction / aim: Visualizing blood flow in small vessel with ultrasound imaging is a useful diagnosis tool for evaluating organ condition. The usage of encapsulated microbubbles that have non-linear response as contrast agents, in combination with pulsing schemes that suppress the linear tissue signal can improve the sensitivity and specificity of ultrasound flow imaging. However, the nonlinear propagation when ultrasound travels through tissue and microbubble clouds reduce the contrast to tissue ratio (CTR) [1]. Moreover, tissue motion causes imprecise re-combination of the multi-pulse transmission scheme, impairing image quality [2]. We have previously developed an independent component analysis (ICA)-based clutter filter for high framerate imaging, which performed better than the widely used singular value decomposition (SVD) filter in presence of fast tissue motion, although it still relied on an SVD pre-filtering step. In this study, we examined the performance of ICA filter in combination with power modulation (PM) pulsing schemes as pre-filtering step to detect flow signal in the presence of nonlinear propagation and motion. Methods: An in-vitro experiment was performed on a flow phantom (polyvinyl alcohol based with background scattering particles) consisting a compartment on top, and a 1 mm diameter channel within the phantom. To obtain ultrasound images in the presence of nonlinear propagation through a microbubble cloud, diluted phospholipid-coated microbubbles were put into the phantom’s top compartment. The microbubbles of interest were infused with a perfusion pump through the channel below the compartment. The pump was turned off just before the acquisition to have very low flow, mimicking microvessel perfusion. A P7-4 ultrasound probe was attached to a linear stage and partly submerged in the microbubbles-filled chamber. A Verasonics Vantage 256 performed the PM imaging. To emulate rigid tissue motion, the probe was attached to a linear stage and moved axially during acquisition, with peak velocities of 35mm/s. Results: The results showed that the combination of PM and ICA filter achieved up to 6dB CTR improvement during fast motion, compared to only PM. This improvement could be significant for myocardial perfusion imaging, where rapid tissue motion, slow microbubble flow, and nonlinear propagation through clouds of microbubbles are expected.
An ultrasound-based modeling framework for the assessment of peripheral arterial disease
Milan Gillissen, Maaike Romme, Arjen van der Horst, Marc van Sambeek, Frans van de Vosse, Richard Lopata
Abstract: Cardiovascular diseases (CVDs) are a leading cause of death, representing an estimated 32% of all global deaths in 2019. It is a group of diseases that involve the heart and the circulatory system, including peripheral artery disease (PAD) which affects over 230 million people worldwide. With atherosclerosis the most common underlying mechanism, lower extremity peripheral arterial occlusive disease (PAOD) is an increasingly common condition in all countries. Currently, the length of the occlusion is used as criteria for endovascular therapy or surgical bypass [1]. However, main challenges of endovascular therapy include the durability of stents in the femoro-popliteal region where the artery is very mobile and the more typical restenosis, which can occur within six to twelve months [2]. Several studies have shown that finite element analysis (FEA) can predict restenosis regions in different arteries [3, 4]. Despite promising results, these patient-specific models are not used in the clinical setting yet. The current computed tomography (CT) or magnetic resonance (MR) approaches suffer from several drawbacks, e.g. ionizing radiation and high cost. Moreover, patient-specific material properties are unavailable in these approaches. We propose a framework based on time-resolved two-dimensional ultrasound (US), which tackles the aforementioned disadvantages as well as acquire the vessel’s motion during the cardiac cycle. The most important drawback of US, besides the low contrast compared to CT and MR, is the limited field-of-view (FOV). A new screening method is proposed in this study which is based on multiple sweeps of the entire upper leg using a two-dimensional ultrasound probe in combination with a probe-tracker. An in-house developed framework is demonstrated, which segments US images automatically, and meshes the geometries obtained for computational fluid dynamics analysis. First results show the feasibility of the technique in healthy volunteers and patients with PAOD. In the future, these models will be used for intervention planning, where a procedure can be applied to the ’virtual patient’ and the outcome and long-term effects can be predicted thereby supporting clinicians in their daily decision-making concerning patients with peripheral arterial disease.
Identification of endarterectomy plaques composition using blind unmixing of multispectral photoacoustic images
Camilo Cano-Barrera, Min Wu, Marc van Sambeek, Richard Lopata
Abstract: Vulnerable carotid atherosclerotic plaques are a major cause of ischemic stroke. Vulnerability of plaques alludes to a high risk of thrombus formation, which can occur due to plaque rupture. This event is associated to specific characteristics of the morphology and composition of the lesions. Therefore, for a better understanding of plaque progression, a comprehensive characterization of their composition is paramount. Multispectral photoacoustic imaging (MSPAI) is a non-invasive technique that can discern the constituents of vascular tissue from their characteristic optical absorption, providing essential information for vulnerability assessment. In this study, we demonstrate the use of blind unmixing techniques on MSPAI to address different types of plaques. We performed ex-vivo MSPAI on carotid endarterectomy samples (n=10) placed in a water tank that allows multi-perspective acquisition. Images were acquired for wavelengths from 500 nm to 1300 nm (in steps of 5 nm), ensuring that the main features of the different chromophores in the plaques were measured. To analyze the spectral modulation at every pixel of the images, we adapted a blind endmember determination algorithm based on the iterative minimization of a target function. It compares multiple sets of endmembers and abundance over different regions on the image and estimates the better hyperspectral imagery representation. The abundance maps define the segmented areas of the sample, and the endmembers are used to classify the material. Results show that blind unmixing of MSPAI can differentiate constituent regions within a sample and provide the endmembers for further analysis. Moreover, spectra do not require an extensive fluence correction to identify multiple materials. PA plaque unmixing results generally agree with the corresponding histological staining. Observable materials are lipids, fibrous tissue, smooth muscle cells, and hemorrhages. Finally, wavelength optimization was used to determine a set of six wavelengths for future acquisitions, which would provide sufficient input for the blind unmixing of the tissues of interest using our acquisition setup.
Detection of freezing of gait and fear of falling in daily live
Juan Delgado Terán, Tjitske Heida, Richard van Wezel, Laurens Kirkels, Robin Jansen
Abstract: Freezing of gait (FoG) has been defined as a brief episode where the subject cannot move forward despite the intention to walk. FoG often leads to balance impairments and constitutes a frequent cause of falling in patients with Parkinson’s Disease (PD). FoG is frequently in coexistence with non-motor symptoms such as depression, anxiety, and Fear of Falling (FoF), which are the strongest predictors of low Quality of Life. Assessment of most PD symptoms is currently performed via questionnaires, and this might lead to biases. However, most people with PD are unable to correctly identify FoG, leading to underdiagnosis. Accurate and continuous evaluation of FoG in home environments using wearable sensors data is critical to diagnose and adequately manage FoG. However, most systems have failed to deliver the same performance in home environments because algorithms created in the laboratory do not address the large variability of human behavior seen in daily life (free‐living conditions). Our main goal is the detection of FoG and FoF using Machine Learning algorithms including physiological sensors (PPG (e.g., heart rate variability), EEG, and skin conductivity) and wearable movement sensors (pressure-measuring insoles and inertial measurement units (IMU’s)) under semi‐controlled, semi‐free living, and free‐living conditions. The configuration of the sensors placed on the person with PD is essential to predict FoG and FoF. Moreover, the obtained dataset will be used for training of novel algorithms aimed at investigating the influence of different types, numbers, and locations of sensors on the performance of FoG and FoF detection. Moreover, with the optimized measurement configuration and machine learning algorithms we aim to detect and even predict FoG episodes in semi-controlled and free-living conditions.
Multivariate Recurrence Plots for the analysis of gait
Joël Karel, Elena Heinze, Ralf Peeters, Pietro Bonizzi
Abstract: Recurrence Plots (RP) are a tool for investigating and visualizing recurrent behaviour in dynamical systems. RP were originally designed for dealing with univariate signals. However, when measuring gait, typically three-axial accelerometers are used. In such a case it may be more convenient to create a multivariate RP instead of creating a separate RP for each of the accelerometer signals. A multivariate RP takes into account the redundant information of all given signals simultaneously and combines it in the same plot. In this respect, it is reasonable to think that a multivariate RP should be better equipped to capture the recurrent behaviour of a system than its univariate counterpart, when more signals are available from the system. To test this hypothesis, we analysed 20 seconds of gait of three healthy controls and of one patient from the Long Term Movement Monitoring Database from Physionet. For the univariate RP, the vertical of the accelerometer was used as this is the direction in which the gait is expected to reflect the most. For the multivariate RP, all three axes of the accelerometer were employed. This is illustrated below with an example from a healthy subject, where from left to right the 3D acceleration signals, the univariate and the multivariate unthresholded RPs are provided. It can be observed that the unthresholded multivariate RP expresses the recurrent behavior more strongly. We also observed that disturbances (like stumbles) become vividly expressed in multivariate RP. Hence, this explorative study gives indications that the use of multivariate RP may be beneficial for the analysis of 3D accelerometer signals.
Home-monitoring of cancer=related fatigue in breast cancer patients
Kim Wijlens, Lian Beenhakker, Annemieke Witteveen, Sabine Siesling, Miriam Vollenbroek, Christina Bode
Abstract: Background Yearly, over 15,000 women are diagnosed with breast cancer in the Netherlands. Earlier detection and better treatments have improved survival. However, there is a growing group of women who experience long-term effects of cancer and its treatment. Cancer-related fatigue (CRF) is the most reported health problem, which can lead to a significant decrease in quality of life but is still underdiagnosed and undertreated. On average, CRF treatments are effective in research settings, but not for all patients. Monitoring of fatigue severity and impact of fatigue on quality of life and social participation of the individual patient is needed to find the most beneficial personalized treatments. Therefore, the aim is to develop a toolkit for home-monitoring that allows personal treatment advice for CRF. Method To determine the relevant domains for holistic monitoring in daily-life, literature research was performed, semi-structured online interviews with healthcare professionals and group interviews with patients from four clinical institutions were held. A funnel approach was used to determine the content of the domains, starting with literature search, followed by expert judgment, and consultation of patient advocates. Wearables, apps, and experience sampling methods were considered if validated in Dutch and with breast cancer patients. Where applicable, questions using Likert-scales were selected using the highest discrimination parameter value. Results The home-monitoring toolkit consists of a selection of questions and a wearable to assess the health status of patients. Starting the toolkit with onboarding questions which will recommend the relevant domains for the individual patient. The domains are CRF dimensions (physical, cognitive, and emotional), general CRF, limitation in functioning (social, relational, and work), day pattern (including activity and sleep), personality and coping style. Conclusion A first toolkit for home-monitoring of CRF was developed. After testing the usability and feasibility of the toolkit, we will pilot the toolkit and propose a personal treatment advice. This advice will be validated with patients and healthcare professionals. The toolkit will be integrated within a personal health environment for easy access and to enable sharing collected home-monitoring data and treatment advice with relevant healthcare professionals.
Biomechanics in the wild: A step towards validation of a wearable system for human movement analysis
Huawei Wang, Akash Basu, Guillaume Durandau, Massimo Sartori
Abstract: The current gold standard measuring setup for human movement analysis is the optical motion capture + force plate system used in the locomotion/rehabilitation labs. However, this system requires participants to stay in a small, restricted capture space which limits the possible motion types as well as the duration of measurements. A wearable measurement system is promising to overcome this issue but has been claimed to be inaccurate. This study aimed to build a wearable system, based on the latest commercial products (inertial measurement units (Xsens, Enschede, NL) + pressure insoles (Moticon, München, GE)), that has the ability to measure both kinematic and kinetic motion data and, in addition, to test its accuracy by comparing it with a gold standard system (optical motion capture (Qualisys, Gothenburg, SE) + instrumented treadmill (Motek, Houten, NL)). An experiment was designed and conducted (ethics approved by University of Twente), with 12 participants (age: 25.4±2.7, height: 173.2±7.6cm, weight: 68.1±9.0kg) wearing the wearable measurement system and performing 13 daily movement types, from slow walking to fast running, combining jumping, squatting, lunging and landing, in the capture space of the optical motion capture system. Then, the recorded motion data were post processed to obtain the information of joint angles, joint torques and ground reaction forces (GRFs) for each movement type. From the comparison results, high correlation and low root mean square values were found between these two systems in measured joint angles for all lower body joints (hip, knee, ankle) in all motion types (Pearson's r = 0.929). Vertical GRFs also showed a high degree of similarity (Pearson's r = 0.954). Whereas high correlation was only seen at the ankle joint in calculated joint torques (Pearson's r = 0.917). Knee and hip joint torque contained significant differences (Pearson's r = 0.172 and 0.591, respectively), mainly due to the lack of horizontal GRFs in the wearable measurement setup. Due to the differences in GRF measurement principles, the centre of pressures (CoP) of two systems were not comparable. Only the similar changes in trends were found. In conclusion, the established wearable motion measurement system can provide accurate joint angles, vertical GRFs, as well as joint torques at the ankle joint. As a side outcome, a completed and synchronized daily movement dataset (raw & processed) with both the optical (gold standard) and wearable measurement systems was generated. Future work includes estimating the horizontal forces in the wearable system to achieve accurate joint torques at the knee and hip joints.
Novel technical calibration of ultra-wideband sensors: A step towards accurate and stable human movement analysis
Vinish Yogesh, Lisanne Grevinga, Jaap Buurke, Peter Veltink, Chris Baten
Abstract: Combining Ultra-wideband (UWB) and Magnetic Inertial Measurement Unit (MIMU) sensors applying data fusion methods could deliver a more accurate and stable full-body kinematics assessment. Still, this approach hasn’t been explored, or used, for human movement analysis. Instead, they only have been applied in single-sensor pedestrian tracking and team sports player tracking [5], and localization. Accuracies in position estimation have been found ranging from 7 to 51cm [1-5], which is not sufficient for clinical movement analysis. Successful translation of this integrated UWB-MIMU system in clinical movement analysis would be possible by improving the UWB ranging estimates accuracy, which in turn increases the position estimate accuracy, resulting from the MIMU/UWB data fusion. Yogesh et al., [5] found the UWB ranging to typically have a large systematic bias error, a negligible gain error, and a residual random error. It is expected that adding a calibration procedure for the systematic error components could increase the UWB ranging accuracy (and by extension the data fusion position estimate accuracy). Therefore, this research aims to develop and validate an algorithm for the efficient calibration of the UWB sensors. A technical calibration protocol and method are proposed to eliminate the systematic errors of the UWB sensors. Non-linear equations relating UWB sensor ranging output with actual distance, including a bias error term were derived for each pair in a set of 4 UWB sensors. A secondary unknown in these equations was the actual internal reference point for ranging for each UWB sensor type. A set of these equations for multiple data samples are then solved by a constrained non-linear optimization procedure. This optimization is validated initially with simulated data since the systematic parameters could not be measured in a practical scenario. Simulated data results show an error of 0.94±1.18 cm. This indicates that the developed optimization method is stable and can accurately determine the systematic parameters. However, this has to be further studied and evaluated in a practical scenario. Currently, experimental data is evaluated from UWB/MIMU combination sensor system, custom-developed for short-range on-body use. References 1. Feng, D., et al., Kalman-Filter-Based Integration of IMU and UWB for High-Accuracy Indoor Positioning and Navigation. IEEE Internet of Things Journal, 2020. 7(4): p. 3133-3146 2. Li, X., Y. Wang, and K. Khoshelham, UWB/PDR Tightly Coupled Navigation with Robust Extended Kalman Filter for NLOS Environments. Mobile Information Systems, 2018. 2018: p. 1-14. 3. Yoon, P.K., et al., Robust Biomechanical Model-Based 3-D Indoor Localization and Tracking Method Using UWB and IMU. IEEE Sensors Journal, 2017. 17(4): p. 1084-1096. 4. Zhang, H., et al., Cost-Effective Wearable Indoor Localization and Motion Analysis via the Integration of UWB and IMU. Sensors (Basel), 2020. 20(2). 5. Yogesh, V., et al., Automated monitoring of load exposure and coping in the context of performance in volleyball. 8th Dutch Bio-Medical Engineering Conference, 2020.
A least-squares Finite Element Method for regularization of sparse displacement data
Jan-Willem Muller, Hans-Martin Schwab, Marcel Rutten, Marc van Sambeek, Richard Lopata
Abstract: Ultrasound (US) strain imaging and elastography rely on displacement tracking to determine the strain field. Many techniques have been developed to determine the displacement between two US images, e.g. block-matching and optical flow methods. Regularization techniques have been described in literature to increase robustness of strain determination. However, these methods often provide limited possibilities, in which regularization is based on first order derivative terms. Furthermore, finite-difference schemes are often used, which need a relatively fine sampling of the estimated displacement field and lead to many variables to solve for. In this study, a regularization method is proposed that is based on a least-squares finite element method (FEM), which allows for flexible regularization of sparse displacement data. The method minimizes the L2-norm between the regularized displacement field and the block-matched data. Overhauser elements with 3rd order polynomials were chosen to guarantee the continuity of strains. The method allows for inclusion of regularization terms using partial differential equations. In this study the smoothness, incompressibility, and the rotational motion were constrained. k-Wave simulations were performed to validate the method. RF data of a vessel (E = 500 kPa, ν = 0.5, r = 3.5 mm) were simulated using an analytical displacement expression. Furthermore, the method was applied on ex vivo data obtained using a porcine aorta setup. The displacement tracking performance with and without regularization were compared. The root mean square error of the global circumferential strain in the k-Wave simulations was reduced from 0.99% to 0.54%, a relative decrease of 45%. Accurate radial strains could be determined from the regularized data, whereas this was not possible without regularization. The global mean absolute displacement error was reduced from 313 μm to 151 μm, which is a decrease of 52%. The regularization method could reduce the tracking drift from 741 µm to 168 µm after one heart cycle in the ex vivo experiment. The results indicate the method can improve the overall displacement smoothness and reduce drift. The impact on strain resolution could, however, not be determined in the current setup and will be analyzed in heterogenous phantoms.
Unsupervised and Time-Adaptive EEG-based Auditory Attention Decoding
Simon Geirnaert, Tom Francart, Alexander Bertrand
Abstract: When multiple speakers talk simultaneously, a hearing device such as a hearing aid or cochlear implant does not know to which speaker the user wants to attend. As a result, a hearing device user often simply turns off the hearing device, not being able to follow any conversation. Determining the intended auditory attention of a user is thus a crucial task to properly inform the hearing device about which speaker to enhance and which other speaker(s) to treat as background noise. Therefore, we develop auditory attention decoding (AAD) algorithms to provide this information from non-invasive neurorecording signals, such as electroencephalogram (EEG) signals. These AAD algorithms are a fundamental building block in so-called neuro-steered hearing devices. In this presentation, we introduce our work in EEG-based auditory attention decoding. More specifically, we highlight and showcase some of our recent advancements in unsupervised [1] and time-adaptive AAD, taking major steps forward towards practical neuro-steered hearing devices. A general video about our work (in Dutch) can be found here: https://www.youtube.com/watch?v=wCm54ZgcdJs. [2] provides an introductory overview paper on AAD algorithms. [1] S. Geirnaert, T. Francart and A. Bertrand, "Unsupervised Self-Adaptive Auditory Attention Decoding," in IEEE Journal of Biomedical and Health Informatics, vol. 25, no. 10, pp. 3955-3966, Oct. 2021, doi: 10.1109/JBHI.2021.3075631. [2] S. Geirnaert et al., "Electroencephalography-Based Auditory Attention Decoding: Toward Neurosteered Hearing Devices," in IEEE Signal Processing Magazine, vol. 38, no. 4, pp. 89-102, July 2021, doi: 10.1109/MSP.2021.3075932.
High aspect ratio 3D printed electrodes for recording of neuronal activity in brain organoids
Bjorn de Wagenaar, Jean-Phillipe Frimat, Filip Simjanoski, Rob Hendriks, Jeroen van den Brand, Ronald Dekker, Massimo Mastrangeli
Abstract: Culture and analysis of cerebral organoids hold great promise for research on early stage brain development, investigation of various brain diseases, and development of personalized drugs. An essential tool for characterizing organoid activity is the measurement of neuronal activity within the tissue. Although microelectrode arrays (MEAs) have been used extensively to study the electrophysiological behaviour of two-dimensional (2D) neuronal cultures, solutions for electrophysiological analysis within organoid tissue are scarce to date. Several reports showed the design of 3D electrode arrays; however, many proposed solutions have limited height (insufficient to protrude deep into the organoid)[1], are not compatible with commercial read-out systems and in-vitro devices[2] or are hard to fabricate reliably[3]. Therefore, we aim at developing high-aspect ratio 3D electrodes using 3D printing, allowing a high degree of customization towards the biological target tissue. The present study presents the design, fabrication and characterization of a 3D-printed 3D MEA. The 3D MEA consists of 3D printed hollow pillars with integrated microfluidic channels. By loading the fluidic channels and hollow pillars with electrically conductive material (e.g. liquid metal or conductive glue), 3D electrodes for deep tissue electric recording are realized. Our transparent platform will be interfaced to a printed circuit board compatible with the commercial MEA2100 readout system from MultiChannel Systems for electrophysiological analysis. 3D prints were fabricated using an Asiga MAX X27 DLP printer using transparent MOIIN tech clear resin (385 nm). After printing, the channels and pillars were cleaned from uncured resin using 70% IPA and post-cured using UV. Liquid gallium was used to fill the fluidic channels and hollow pillars. The bulk resistivity of the generated 3D electrode including microfluidic conducting track amounted between 0.3 and 2 Ohm (n = 2). Future work will involve (further) electrical characterization of the 3D electrodes, biocompatibility testing of the 3D printed material, and culture of brain organoids on the electrode array.
Neuromuscular control of the wrist in amyotrophic lateral sclerosis
Diederik Stikvoort, Just Plouvier, Boudewijn Sleutjes, Stephan Goedee, Winfred Mugge, Alfred Schouten, Frans van der Helm, Leonard van den Berg
Abstract: Amyotrophic lateral sclerosis (ALS) is a relentless neurodegenerative disorder with ultimately fatal consequences. The expression of symptoms is highly heterogeneous, with large variation in degeneration rates of both upper and lower motor neurons (UMN, LMN)1. As a result, symptomatic motor behavior in ALS originates from a complex interplay of central and peripheral symptoms, including hyperreflexia, spasticity, rigidity and weakness2. UMN symptoms can be particularly hard to detect due to degeneration of all classes of motor neurons in the anterior horn of the spinal cord2. Secondary changes in muscles and connective tissue further alter the limb’s dynamics and the ensuing reflexive behavior3. Here, we present a protocol to quantify wrist neuromuscular control of ALS patients in order to explore their associations with the clinical manifestation of the disease. We plan to recruit 20 ALS patients, excluding participants with a muscle strength of the wrist muscles below an MRC of 3 (Medical Research Council scale, 0-5), presence of active psychiatric diseases such as frontotemporal dementia, concomitant neuropathy, history or presence of brain injury or other cerebral diseases. Reference data will be derived from age-gender matched controls. Wrist perturbations and closed-loop system-identification techniques will be employed to estimate the neuromuscular control of the wrist joint. Participants will perform multiple tasks while supplemented with unpredictable multisine torque perturbations, using a robotic manipulator, to elicit a wide range of neuromuscular control as previously described4¬. Parameterization of the wrist-dynamics using a neuromuscular model provides metrics describing the contribution of muscle- and reflex-dynamics to the observed motor behavior. Clinical metrics will be obtained, including muscle tone of the arm, reflex score of the examined arm, revised ALS Functional Rating Scale score (ALSFRS-R) and the fine motor function subscore (ALSFRS-R items 4-6). Comparable patterns of change in neuromuscular control will be identified through hierarchical clustering of the neuromuscular parameters. The clinical characteristics of these clusters will subsequently be compared. This\ study is the first to explore neuromuscular control in ALS with neuromuscular modeling and system-identification techniques. Quantifying changes related to UMN degeneration may be particularly useful for following disease progression in a clinical or clinical trial setting. 1. van Es MA, et al.. Amyotrophic lateral sclerosis. Lancet 2017;390(10107):2084-2098. 2. Swash M. Why are upper motor neuron signs difficult to elicit in amyotrophic lateral sclerosis? Journal of neurology, neurosurgery, and psychiatry 2012;83(6):659-662. 3. Kamper DG, Schmit BD, Rymer WZ. Effect of muscle biomechanics on the quantification of spasticity. Ann Biomed Eng 2001;29(12):1122-1134. 4. Mugge W, et al.. A rigorous model of reflex function indicates that position and force feedback are flexibly tuned to position and force tasks. Experimental Brain Research 2010;200(3-4):325-340.
Development of non-invasive method to measure the mechanical properties of skin using ultrasound techniques
Zülal Kizilaslan, Marcel Rutten, Frans Van De Vosse, Richard Lopata
Abstract: Studies have shown that measuring the mechanical change of skin can be used to detect skin diseases such as Pressure Ulcer (PU) [1,2]. The objective of this study is developing an experimental setup to estimate the mechanical properties of skin based on quasi-static ultrasound (US) elastography, a technique used to investigate the biomechanical quantities and properties of tissue (such as strain, modulus) non-invasively [3]. For an in vitro feasibility test, a phantom was created, using a 15 weight percent (wt%) polyvinyl alcohol (PVA) solution. For acoustic scattering, 3 weight percent silicon carbide was added to the PVA mixture. Next, a linear array US probe (10 MHz) was used to both compress and image the phantom simultaneously. To create a heterogeneous deformation field and measure the load, a water filled, small diameter balloon, which is attached to a pressure sensor, was positioned between the probe and the upper surface of the artificial tissue. During compression, the hydrostatic pressure that is present inside the balloon and is transferred to the tissue, was recorded. Ultrasound imaging was performed during compression. The displacement field in the vertical direction was estimated by analyzing the radio-frequency (RF) ultrasound data using a 2-D block matching technique and converted into strains. Finite Element Analysis (FEA) combined with mechanical test were adopted to verify the US experiment result. The 3D FE model of the subjects (PVA phantom, water-filled balloon and the probe as indenter) was designed on Abaqus software. The uniaxial tensile test results were used in order to determine the nonlinear, hyperelastic constitutive parameters of the PVA phantom and the balloon. The mesh sensitivity and the proper contact between the domains will be addressed for an accurate FEA.


end %-->