• Home
• Preliminary Program

16:00   Poster Session 1 (Even numbers) 16:00 0 mins Beyond the Nyquist criterium: The effect of sampling frequency and angular velocity variability on orientation estimation from gyroscope data Anne Haitjema, Frank J. Wouda, Kim Sunesen, Jasper Reenalda, Bert-Jan F. van Beijnum, Peter H. Veltink Abstract: Inertial measurement units (IMUs) are frequently used in human movement analysis. A suitable sampling frequency must be chosen for accurate analysis. The Nyquist criterion, which states that the sampling frequency should be at least twice the highest frequency contained in the signal of interest, is often used to ensure accurate representation of continuous-time signals in the discrete-time domain. However, orientation estimation from sampled angular velocity might be subject to additional sources of error. We want to evaluate, both theoretically and empirically, the influence of sampling frequency and multidimensional angular velocity variability on the accuracy of orientation estimation from human angular velocity measurements. Theoretically, relative orientation can be estimated by integrating the angular velocity according to Bortz [1]. The discrete nature of this calculation can introduce errors. Two potential factors influencing the magnitude of this error are the multidimensionally variable angular velocity, and the timestep chosen in the integral. For the empirical evaluation, an experiment will be designed in which 10-15 participants will walk and run whilst instrumented with 4 IMUs (sternum, pelvis and tibias) that output raw sensor data at 1000 Hz. An optical motion capture system will be used as a reference. The different sensor locations and movements will be used to evaluate the effect of the variability in multidimensional angular velocity on orientation estimation. The effect of the sampling frequency on orientation estimation will be evaluating by downsampling the data. Results from a pilot experiment involving walking data collected at 1000 Hz (pelvis and tibia) indicate that decreasing the sampling frequency increases the error in the orientation estimation, even though inspection of the frequency spectrum suggests that the Nyquist criterion is fulfilled. Furthermore, a lower multidimensional variability in the angular velocity seems to lead to a lower absolute error in the orientation estimation. If it is possible to represent a movement around one main rotational axis, then this decreases the error in the orientation estimation [2]. We hypothesize that the estimation error increases when human movements exhibit a greater multidimensional variability in the angular velocity. This work is partly funded by TKI HTSM and Movella Technologies BV. References 1. Bortz JE. A new mathematical formulation for strapdown inertial navigation. IEEE Trans Aerosp Electron Syst. 1971 Jan; 7(1): 61-66. 2. Zandbergen MA, Reenalda J, Middelaar RP van, Ferla RI, Buurke JH and Veltink PH. Drift-free 3D orientation and displacement estimation for quasi-cyclical movements using one inertial measurement unit: Application to running. Sensors. 2022 Jan 26; 22: 956. 16:00 0 mins Thermal cameras for respiration monitoring: comparing the performance of different placements Raquel Alves, Fokke van Meulen, Mark van Gastel, Wim Verkruijsse, Sebastiaan Overeem, Sveta Zinger, Sander Stuijk Abstract: In a polysomnography exam, patients get their vital signs monitored through sensors attached to their bodies. The inconvenience of contact sensors can be partially overcome with the use of thermal cameras. Thermal cameras can monitor respiration motion and flow, by detecting chest movements and temperature variations caused by breathing airflow [1]. It is important to understand how many cameras should be used and where they should be placed. This study used 6 low-cost/low-resolution cameras to collect data from 25 subjects. The cameras were positioned around a bed where subjects were asked to lay in different sleeping positions and breathe at a normal rhythm, either through the nose, mouth or with a nasal cannula. A respiration belt was used as a reference. The respiration signal was extracted from the videos using the algorithm from [1]. It extracts three features from the thermal videos to define a region of interest where the respiration signal is the average pixel intensity over time. The algorithm can use the input from a single camera or a combination of cameras to estimate respiration. To understand how the estimated respiration signal changes when different camera positions are combined, all possible combinations were tested. To evaluate the results, the relative error, precision, and sensitivity values were computed. These last two metrics were computed considering the detection or missed detection of each breath individually. The results show a lower relative error (8.52%) and a higher precision (91.95%) and sensitivity (96.47%) when combining the cameras located behind the headboard with the cameras on the side of the bed. Using the maximum number of cameras does not result in the best outcome. Results also show that breathing through the nose or a nasal cannula often leads to a lower measurement error. There were no significant differences detected between the results of the two sleeping positions tested (supine and lateral). This is the first study that gathers information essential for a conscious placement of thermal cameras for use in future studies and clinical applications. It not only describes the best setup but also gives possible alternatives when considering placement constraints. 16:00 0 mins Modelling the placenta vasculature as a basis for clinical decision support Pascalle Wijntjes, M.B. van der Hout-van der Jagt, W. Huberts, F.N. van de Vosse Abstract: Approximately 10% of all pregnancies are affected by hypertensive disorders, with 3% being affected by the more severe condition of pre-eclampsia. Studies show that multiple pregnancy complications are caused by abnormal placentation leading to placental dysfunction and severe complications. Therefore, it is important to look at the placenta vasculature in more detail. In this study, we will investigate if it is possible to gain more insights into the placenta physiology, i.e., its functioning and remodeling by creating a hemodynamic vasculature model of the placenta. This new knowledge could be used in (hybrid) prediction models for pregnancy complications like pre-eclampsia. A feto-placental vasculature model is created. For this, a space-filling algorithm is used to create the vessel tree within the given geometry, which defines the placenta and its functional segments (cotyledons). The model is assembled by assumptions made in the mathematical model based on literature and clinical knowledge. The resulting modelled placenta realistically mimics the placental vasculature on visual inspection. Further validation requires data that is not yet available to evaluate the vasculature details. However, the current model can be used to evaluate the hemodynamic functioning of the placenta for clinical scenarios. Simulations will be based on questions given by obstetricians. Ultimately, the goal is that this model and its simulations will in the future be able to support clinical decision-making and give a basis for new prediction models for pregnancy complications. 16:00 0 mins Extraction of cardiac-related signal from suprasternal pressure sensor in polysomnigraphy Luca Cerina Abstract: The accurate detection of respiratory effort during Polysomnography (PSG) is critical in the diagnosis of sleep disordered breathing conditions such as sleep apnea. Unfortunately, the sensors used in the clinical routine are invasive or do not capture upper airway dynamics. One promising Extraction of cardiac-related signal from suprasternal pressure sensor in polysomnigraphyalternative is the Suprasternal Notch Pressure Sensor (SSP): a small sensor placed on the skin that detects pressure swings in the thoracic cavity. Besides respiratory activity, the SSP may also pick up small pressure oscillations caused by the pumping heart. Although these are commonly removed as unwanted artifacts, they have potentially informative content regarding cardiac activity. In this study we propose a method to extract cardiac information from the SSP signal. First, the raw signal is filtered to attenuate respiratory frequencies. Then we get robust-to-noise estimates of the heart rate (HR) as local maxima in the signal’s autocorrelation. The frequency search range is determined through a-priori knowledge of HR dynamics and by tracking the temporal evolution of HR estimates. Finally, we tune time-variant filters on the HR to separate respiratory and cardiac signals. The performance in HR estimation is compared with a ground truth extracted from synchronized ECG recordings on a sample of 100 participants undergoing a full single-night PSG, including the SSP sensor, for various suspected sleep disorders. Since the transition to sleep apnea events may hinder our method, we also measured the loss of performance compared to normal breathing. The respiratory signal filtered using our method or a fixed frequency notch filter at 1.6Hz (currently employed) are compared qualitatively. Pooling all the estimates, the Bland-Altman agreement analysis resulted in a linear bias of -0.06bpm with 95% level-of-agreement of 5.09bpm. The coverage of SSP noise-free estimates compared to the ECG was 94.4±2.3%. A paired Wilcoxon Rank-Sum test determined that the error caused by respiratory events is significant across the experimental population, with an average increase of 0.38bpm (interquartile range 1.44bpm). We showed that besides thoracic respiration pressure swings, the SSP sensor contains reliable cardiac information. Our method achieved good results in estimating the HR without additional sensors and unlocked new research activities regarding the extracted cardiac signal. 16:00 0 mins Pilot validation study of inertial measurement units and markerless methods for 3D neck and trunk kinematics Ce Zhang, Christian Greve, Gijsbertus Jacob Verkerke, Charlotte Christina Roossien, Han Houdijk, Juha M. Hijmans Abstract: Background: Musculoskeletal symptoms (MSS) are a major health issue in different kinds of occupations. Surgeons are a group of healthcare professionals who are at high risk for developing MSS, such as neck and low-back pain and even herniation. A passive exoskeleton is a potential solution that can prevent developing MSS. The function of the exoskeleton is to store energy during bending and releasing it during extension and thus reduce muscle force during surgery. To get the input for designing the exoskeleton such as the range of motion (ROM) and desired stiffness, it is necessary to analyse the kinematics of surgeons in the operating room (OR). However, optoelectronic systems are not suitable for the OR. Inertial Measurement Units (IMU) and markerless (DeepLabCut) motion capture methods could be a good alternative for optoelectronic systems in the OR. We aimed to validate IMU and markerless methods against an optoelectronic system as gold standard. Method: Ten healthy young subjects participated in this research. The motion tasks were movements in primarily a single anatomical plane and simulated surgery task. The 3D neck and trunk angle were obtained by the IMU, markerless and Vicon system. Intraclass correlation coefficient [ICC (2,1)], root mean square error (RMSE), range of motion (ROM) difference and Bland-Altman plots were used for evaluating both methods. Results: The IMU-based motion analysis showed good-to-excellent (ICC 0.80 - 0.97) agreement with the gold standard within 2.3 to 3.9 degrees RMSE accuracy during simulated surgery tasks. The markerless method shows 5.5 to 8.7 degrees RMSE accuracy (ICC 0.31 - 0.70). Discussion: The IMU-based method maintains a high level of accuracy even during complex simulated open surgery tasks with larger ROM in the neck and trunk segments. The markerless method cannot provide stable validated kinematic results, because of virtual tracking point occlusion, camera setup and tracking error due to the DeepLabCut algorithm. The present markerless method is not yet sufficiently valid, but it might have the potential for 3D movement analysis in the OR if the camera setup is improved and the model is trained by more data. Therefore, for now it is recommended to use IMU for the kinematic analysis of head and trunk motions in the OR to provide input for ergonomic interventions. 16:00 0 mins Do in-vitro neural networks predict ? Martina Lamberti, Shiven Tripathi, Sarah Marzen, Joost le Feber Abstract: Several studies have suggested that memory and prediction are crucial neuronal functions directing our actions. When an external input is perceived, its immediate registration is observed as short (seconds) lasting activity changes, which can be seen as short-term memory, whereas long-term memory refers to connectivity changes on time scales of minutes to hours. Prediction can be defined as the ability to reduce uncertainty on future sensory input, and has been hypothesized to depend on memory, particularly on short-term memory. Recent work showed that retinal cells predict visual stimuli, but it is yet unknown whether prediction is a general capability of neuronal networks. Here, we determined whether in vitro neural networks predict external stimuli, and if prediction depends on (short-term and long-term) memory. We used multi electrodes arrays (MEAs) with 59 recording electrodes on which rat primary cortical neurons were plated. We electrically or optogenetically stimulated for 20 hours with interstimulus intervals (ISIs) taken from a known distribution. Repeated electrical stimulation at one electrode has been shown to induce significant long-term connectivity changes, interpreted as long-term memory traces. In contrast, optogenetic stimulation did not. We used mutual information to quantify to what extent recorded activity reduced the uncertainty on future stimuli (MIfuture; prediction), or recent past stimuli (MIpast; short-term memory). Activity provided significant information on past stimulation, indicating that stimulus responses clearly deviated from spontaneous activity. MIfuture reflected the distribution of ISIs, suggesting that it largely depended on stimulus responses. This was confirmed by masking stimulus responses which largely reduced MIfuture. Throughout 20h of stimulation, MIfuture linearly depended on MIpast. However, during electrical stimulation this dependency on short-term memory significantly decreased with time, suggesting that other features gradually took over. Optogenetic stimulation, in contrast to electrical stimulation, did not induce long-term memory traces and showed unchanged dependency of MIfuture on MIpast. We conclude that random neuronal networks predict future stimuli, predominantly based on short-term memory of past stimuli. With the induction of long term memory traces, the dependency on short-term memory becomes less dominant. 16:00 0 mins An in vitro and in silico neuronal network model to unravel genetic encephalopathies Nina Doorn Abstract: The brain is still the least understood organ in the human body, and there is a lack of understanding of many neurological disorders. It is vital to develop efficient models of these disorders to uncover disease mechanisms and develop effective therapeutic strategies. Studying the brain in vivo is complicated by the limited possibility to control and invasively measure neuronal functioning. Neuronal networks derived from human induced pluripotent stem cells (hiPSCs) can potentially overcome these issues, as they allow to study physiological and pathological network behaviour in vitro with a patient-specific genetic signature. These networks, derived from healthy subjects and patients, are spontaneously active and show robust and replicable electrophysiological signatures that can be observed using multi-electrode arrays (MEAs). Various genotype/phenotype correlations have been established with this method [1]. Despite these advances, the identification of cellular and synaptic mechanisms underlying abnormal phenotypes remains challenging, as this is not trivial to deduce from the networks’ electrical phenomenology. Therefore, we developed a biophysical in silico model that faithfully reproduces the activity of healthy control neuronal networks on MEAs. To illustrate the potential of the model to test hypotheses and predict underlying disease mechanisms, we studied excitatory neuronal networks derived from a Dravet Syndrome (DS) patient. DS is a highly studied, severe infantile epileptic encephalopathy, caused by mutations in SCN1A encoding part of the voltage-gated sodium channel NaV1.1. It remains a paradox how impaired sodium currents result in epilepsy, especially in excitatory neurons. Our in silico model revealed that sodium channel dysfunctions were insufficient to transition from a model of healthy networks to a model resembling the in vitro DS network behaviour, and that additional alterations were needed. In particular, the model predicted the influence of reduced after-hyperpolarizing currents and synaptic strengths in DS neuronal networks. We subsequently substantiated these in silico generated predictions in vitro, providing new hypotheses about cellular mechanisms at play in the DS neuronal networks. This illustrates the potential of our in silico model to identify important mechanisms that can then be investigated in vitro in a targeted and efficient manner, expanding our understanding of healthy and patient-derived neuronal networks. 16:00 0 mins Regulating microclimates in biomedical product interfaces by surface engineering Hanneke Reuvekamp, Dave Matthews, Edsko Hekman, Emile van der Heide Abstract: Alongside the increasing amount of personal healthcare devices being used, also more medical-device related injuries (MDRI’s) are experienced at the interface. To be able to answer the research question of how these injuries can be prevented through intelligent product design, a systems approach showed that the predominant underlying cause is prolonged contact with an unfavourable microclimate, comprising the effects of humidity, temperature and airflow. In general, managing MDRI’s is hard from dermatology and engineering practice: It shows that up till now it is a process of trial-and-error and the focus is on reducing, rather than preventing. In the end, the use of these products, the time needed as well the treatment of other consequences such as pressure ulcer development or dermatitis are paired with significant costs. Surface engineering is well known to further modify and possibly enhance the functionality and performance of a product, including the thermo-mechanical interactions taking place at the skin-medical device interface. In order to be able to reduce the trial-and-error time and create personalised solutions for microclimate regulation through surface engineering, steps towards research and design must be taken. Engineers and healthcare professionals report that future work should focus on creating design maps supporting the engineering process in which microclimate regulation mechanisms are designed towards an optimised extent. This research will address the role of surface engineering in microclimate regulation of personal healthcare devices in prolonged contact with the skin by introducing a design map for the development of interfaces managing both the thermal and moisture component. This will be done with respect to the correlation of texture features, material properties, and product characteristics. The presentation will link knowledge from dermatological, biomechanical and engineering perspectives and current and future microclimate regulation strategies from a systems approach as a first step towards the proposed design map. 16:00 0 mins A biomedical approach using RTLS for quality assessment in trauma care Igor Paulussen, Alyssa Venema, Anne-Marie van Oers, Mike Bemelman, Lise van Turenhout, Gerrit Noordergraaf Abstract: Main research question: Workflow management in trauma care has been associated with higher efficiency and effectivity. Carefully standardized practice, particularly in teams with large numbers of participants per discipline, are a golden standard. However, monitoring workflow, professional activities and movement within the trauma bay remain exceptionally difficult. Can a Biomedical Engineering (BME) approach help? The Elisabeth-TweeSteden Hospital (ETZ) recently investigated the stoke pathway workflow using a Real Time Location System (RTLS). This innovative system registers the location of patient, nursing and medical staff every moment during the care pathway. Granularity was low at “room” level. We will assess this bio-engineering approach in its ability to offer insights into a team approach using a high granularity setting in a trauma bay around a trauma stretcher during actual acute care admissions. Research method: The ETZ workflow for acute care admissions is per-protocol, presumably leading to a strongly coordinated, reproducible, rapid decision making and high quality process. This, however, has not been validated and ad hoc verbal feedback suggests potentially large protocol interpretations and deviations. In the proposed study the RTLS granularity will be higher than previously described and combined with output from the real time Electronic Medical Record (EMR). Multiple virtual, small, designated work areas have been identified within the trauma bay, allowing recognition of movement in- and outside each task area. For privacy purposes, RTLS Tags will be incorporated in the discipline-coded vests. The patient will receive a tagged bracelet. Results & discussion: We expect circa 100 cases per month. Outcome measures are: a) adherence to the workflow protocol and cause-effect relationships, corrected for trauma severity; b) relationships between frequency of out-of-area movement and time/quality; c) relationships between team assessment of their own workflow adherence versus RTLS/EMR demonstrated adherence; d) ability of RTLS (tags) to support the granularity within a trauma bay. Conclusions: This study will landmark logistic and professional quality in a high granularity setting showing the value of integrating engineering in care pathways. 16:00 0 mins Image quality and validity of handheld echocardiography for stroke volume and left ventricular ejection fraction quantification Frederique de Raat Abstract: Background: Bedside quantification of stroke volume (SV) and left ventricular ejection fraction (LVEF) is valuable in hemodynamically compromised patients. Miniaturized, battery operated handheld ultrasound (HAND) devices are now available for clinical use. However, the performance level of HAND devices for quantified cardiac assessment is yet unknown. The aim of this study was to compare the validity of HAND measurements with Standard Echocardiography (SE) and three-dimensional echocardiography (3DE). Methods: Thirty-six patients were scanned with HAND, SE and 3DE. The image quality of HAND and SE was evaluated by scoring segmental endocardial border delineation (2=good, 1=poor, 0=invisible). LVEF and SV of HAND was evaluated against SE and 3DE using correlation and Bland-Altman analysis. HAND and SE datasets were analysed automatically with Auto Strain. The 3DE dataset was analysed with the Dynamic Heart Model. Results: The correlation, bias, and Limits Of Agreement (LOA) between SE and HAND were 0.68 [0.46:0.83], 1.60% [-2.18:5.38], and 8.84% [-9.79:12.99] for LVEF, and 0.91 [0.84:0.96], 1.32 ml [-0.36:4.01], 15.54 ml [-18.70:21.35] for SV, respectively. Correlation, bias, and Limits Of Agreement (LOA) between HAND and 3DE were 0.55 [0.6:0.74], -0.56% [-2.27:1.1], and 9.88% [-13.29:12.17] for LVEF, and 0.79 [0.62:0.89], 6.78 ml [2.34:11.21], 12.14 ml [-26.32:39.87] for SV, respectively. The image quality scores were 10.49 ± 1.7 for the apical four chamber views for the SE dataset and 9.42 ± 2.0 for the HAND dataset (P < 0.001). Conclusion: Clinically acceptable accuracy, precision, and image quality was demonstrated for HAND measurements compared to SE. Also, LVEF quantification with HAND was interchangeable with 3DE. 16:00 0 mins Towards personalized immersive virtual reality (IVR)-based stroke neurorehabilitation: a user-centered design approach Salvatore Luca Cucinella, Joost de Winter, Heike Vallery, Laura Marchal-Crespo Abstract: Immersive Virtual Reality (IVR)-based rehabilitation allows people suffering from a stroke to train their paretic limbs in controlled virtual environments with the aim of regaining their cognitive and motor functions [1,2]. Compared to conventional VR-based rehabilitation using computer screens, IVR using head-mounted displays can be tailored to the patient's physical and psychological constraints and goals. Thus, in IVR, patients training in a first-person perspective experience a more naturalistic and controlled representation of reality [3]. This visualization technology aims to stimulate patients physically and mentally, thus promoting neuroplasticity; however, its effectiveness depends on multiple intrinsic and extrinsic factors playing a crucial role in the therapy outcome, such as motivation [4]. Using a User-Centered Design (UCD) approach and employing the Double-Diamond model as the research framework [5], we aim to investigate these factors to design a suitable virtual training environment and - eventually, to evaluate their impact on the effectiveness of (and adherence to) IVR-based neurorehabilitation among people who suffered a stroke. User observations and semi-structured interviews with ten rehabilitation experts from Rijndam Revalidatie, Rotterdam, will help us define the patients' experience and gain insight into the therapy treatment. This research aims to produce knowledge regarding the role and responsibilities of the healthcare professionals involved in the care of patients, the tools used for assessing the patients' motor and cognitive abilities, and the intrinsic and extrinsic factors influencing the rehabilitation outcomes. In a following co-creation session, participants will be involved in activities to generate new ideas on the requirements of virtual training environments, thus shaping the future of technology-driven rehabilitation. 16:00 0 mins Capturing uncertainty in overnight sleep statistics using automatic scoring Hans van Gorp, Merel van Gilst, Pedro Fonseca, Ruud van Sloun, Sebastiaan Overeem Abstract: Objectives: Sleep staging is a crucial but labour-intensive diagnostic tool, compelling the use of automated machine learning models. These models are trained on scorings made by humans, which have imperfect agreement, resulting in uncertainty about the correct scoring. Recent developments in automatic sleep staging, such as hypnodensity, try to capture this uncertainty on an epoch-by-epoch basis. However, these methods are inadequate to capture the uncertainty of overnight sleep statistics. Here, we propose U-Flow, which given a single input PSG, outputs a desired number of plausible hypnograms, mimicking the diversity seen in human scorings. Methods: The model was trained with 529 recordings of the Stanford Sleep Cohort (SSC; single scorer), all 80 recordings of the Dreem Open Datasets (5 scorers), and all 110 recordings of the Institute of Systems and Robotics dataset (2 scorers). The validation set consisted of the remaining 132 recordings of SSC (single scorer). The hold-out test set comprised 70 recordings of the Inter-Scorer Reliability Cohort (6 scorers). We compared U-Flow to a hypnodensity-based U-Net model of similar computational complexity. For each input PSG, both models output 100 hypnograms, which were compared to those of the 6 experts. Accuracy and kappa were calculated through majority voting. Additionally, 9 commonly-used summarizing sleep statistics were calculated from each hypnogram, including total sleep time, time spent in each stage, sleep onset latency, REM onset latency, and number of awakenings. Each sleep statistic for each recording was modelled as a normal distribution, resulting in a mean and a variance, where the latter expresses the uncertainty. The distance between the distributions of the experts and the model predictions was evaluated using the Kullback–Leibler divergence. Significance was tested through ANOVA. Results: U-Flow was found to better predict the distribution of overnight sleep statistics by a human panel for 8 out of 9 parameters (p<0.05), while not sacrificing accuracy and kappa. These were respectively 80.0% and 0.71 for U-Net compared to 82.7% and 0.74 for U-Flow. Conclusions: U-Flow outperforms a hypnodensity-based U-Net model of similar computational complexity, in its ability to capture the uncertainty of overnight sleep statistics, as well as accuracy and kappa. 16:00 0 mins GoodUseT, A safety II methodology to share experiences around use of medical technology among healthcare professionals Herke Jan Noordmans, Wendela de Lange, Kim de Groot, Bas de Vries, Jaap Trappenburg, Jan-Bernhard Broekhuis, Lisette Schoonhoven Abstract: When utilizing medical devices, mistakes can be made that may lead to an injury to the patient. For about 15 years, hospitals have set up incident and calamity processes to structurally track this information and try to conceive measures to reduce the chance of recurrence or extent of a possible injury. This has worked well for years, but now hospitals are beginning to realize that the low-hanging fruit has been harvested. Even more measures lead to additional registration burden or reduced efficiency, and improving the care process in one place often makes the care process at another place more cumbersome or even riskier. Hence, (high-risk) organizations worldwide are looking at what they can learn from practices that normally go well. This Safety II approach looks at variations in daily work and makes them discussable. Perhaps good practices in one department can also be useful for another department, although they often have to be adapted for that specific situation. In this study, we applied the Safety II approach to nursing care in the pulmonary and pediatric ICU departments at the UMC Utrecht. We compared work in practice (Work as Done) with what was described in protocols and work instructions (Work as Imagined) through observations and interviews. This was done for the process around the feeding pump, antibiotics administration via infusion and ventilation. User variations regularly emerged in which one user handled things differently (under certain circumstances) than another. Identifying and discussing these variations was felt to be very valuable, especially since this discussion was almost never held. At the end of 2022, we want to further analyse and report the results. We also want to test self-developed WhatsApp software to let users of a particular healthcare technology exchange experiences with each other across the boundaries of a department. 16:00 0 mins Neuronal stimulation or hypothermia protects neuronal network functioning during hypoxia Eva Voogd, Sara Pires Monteiro, Jeannette Hofmeijer, Monica Frega Abstract: Cerebral ischemia is a pathological condition caused by lack of blood flow to the brain, where low oxygen and glucose levels have detrimental effects on neuronal functioning and viability. So far, effects of neuroprotective therapies derived from animal models could not be translated to patients. A human disease model holds potential to improve identification and translation of new treatment strategies. Here, we created a human-induced pluripotent stem cell (hiPSC) derived neuronal model to investigate key pathomechanisms in cerebral ischemia and identify potential treatment strategies. In this model, we can assess neuronal activity with micro-electrode arrays and synaptic connections, as well as cell survival, with immunocytochemical techniques. We showed that hiPSC-derived neuronal networks have decreased spontaneous activity and decreased synchronous activity during and after hypoxia, indicating decreased network connectivity. In line with these findings, our results showed decreased synaptic connections and, subsequently, decreased cell survival during hypoxia. We used optogenetics to generate hiPSC-derived neurons responsive to light to investigate whether neuronal network stimulation would be neuroprotective during and after hypoxia. Our results showed that baseline levels of spontaneous and synchronous neuronal network activity were maintained with optogenetic stimulation during and after hypoxia. In addition, we investigated effects of hypothermia. The results showed preservation of synapses during hypoxia in hiPSC-derived neurons subjected to hypothermia. We conclude that hiPSC-derived neuronal networks are vulnerable to hypoxia, with synaptic failure being a key pathomechanism. We also conclude that neuronal network stimulation as well as hypothermia hold potential to prevent loss of network functioning during and after hypoxia. 16:00 0 mins Pushing boundries – exploring the importance of jugular venous distention as a sign of tention pneumothorax. a Bio-Engineering solution for clinical research introducing really dangerous conditions Tristan van Leest, Matthijs Vogels, Igor Paulussen, Bart Spruijt, Gerrit Noordergraaf Abstract: Introduction: Tension pneumothorax is a potentially lethal injury by leading to respiratory and circulatory arrest. As a simple pneumothorax may progress to tension, clinicians must identify and treat it early. In trauma courses signs of tension pneumothorax are respiratory distress, hypoxia, tracheal deviation and particularly (early) jugular venous distention (JVD). However, disagreement exists in JVD’s timely (early/late) appearance and its relationship to volume status. We aim to identify the limits of unilateral pressure thresholds to better understand the role of jugular venous distention as a symptom in tension pneumothorax. This will show if JVD is a clinically important entity in diagnosing (early) tension pneumothorax allowing recognition of the danger for imminent collapse. Methods: Elective videoscopic thoracostomy (VATS) patients would be a suitable population as animal models (i.e. swine) are less suitable due to their anatomy. However, introducing enough (air) pressure to induce circulatory collapse is potentially an ethically challenging form of clinical research. While resolution of the circulatory collapse should be uncomplicated, inclusion seems daunting. Therefore, we explore a bio-engineering approach using an ex-vivo digital model based on the Donders model for CPR. The extended model should simulate the transmural effect of pressure on the central veins, the effect of mediastinal lateralisation on lumen collapse (vascular twisting), in addition to the effect of global increase in intra-thoracic pressure on the vasculature and right heart. These are deemed contributing factors to the circulatory collapse (Pulseless Electrical Activity). Nonetheless, finding behaviour variables for the model is difficult and validation more so. Results: We hypothesise that increasing the unilateral, intra-thoracic pressure, will non-linearly diminish venous return. This should result in venous distention of the right external jugular vein as Niemann’s valve will not operate, plus increased vessel diameter of the internal jugular veins above a certain pressure threshold. Discussion & Conclusion: Clinical limits suggest bio-engineering approaches to address relevant clinical dilemma’s. Jugular distention is only useful as a marker if it has a consistant and timely appearance. We believe that this will advance not only current daily clinical practice, but also future education of medical professionals. 16:00 0 mins Smart module for removal of protein bound uremic toxins for artificial kidneys Jeroen Vollenbroek, Fokko Wieringa, Karin Gerritsen, Joachim Jankowski, Lucas Lindeboom, Rosalinde Masereeuw, Leonard van Schelven Abstract: Most End Stage Kidney Disease (ESKD) patients rely on hemodialysis to remove toxins from their blood. During hemodialysis, toxins smaller than 20-45 kDa are filtered out of the bloodstream into the dialysate. Some small toxins, the Protein Bound Uremic Toxins (PBUTs), are difficult to remove since they are bound to large proteins (e.g. albumin ~66 kDa) [1]. Whereas healthy kidneys can actively and very efficiently secrete these PBUTs, hemodialysis only removes the unbound fraction [2]. PBUTs-accumulation is implicated with injury to the heart, blood vessels, brain and nerves. Simply increasing filter pore size is not an option since the binding blood proteins are vital and should not be removed [1]. The Multi-compatible Implantable Toxin Removal Augmentation Module (MI-TRAM) may greatly enhance PBUT-removal by loosening PBUTs from blood proteins, thereby increasing the free fraction that will be filtered out. A consortium from RWTH Aachen, UMC Utrecht, Utrecht University and IMEC is working on this. High strength, high frequency electromagnetic (EM) fields will be used to shake the electrostatic bonds between blood proteins and PBUTs [3-5]. In vitro experiments by prof. Jankowski, already demonstrated significantly improved PBUT removal using bulky equipment [3]. The MI-TRAM chip, developed by IMEC, is a miniaturized version of this system, compatible with wearable, portable or implantable artificial kidney systems. The chip integrates sensors to monitor hematocrit level, total body water and body temperature. We have built a setup with in-tube stainless steel connectors applying the EM fields across the dialysis membrane and are developing an electrical model for the response and behaviour of the system. In the near future, we will perform in vitro dialysis experiments with the MI-TRAM chip for performance testing, calibration, optimization of field strength and frequencies. Both conventional and novel dialysis membranes will be tested for PBUT removal efficiency enhancement. After successful in vitro testing, we will proceed with in vivo efficacy and safety experiments in a uremic large animal model [6], followed by clinical trials. MI-TRAM has the potential for a rapid clinical introduction of this technology to the benefit of ESKD patients. References: [1] Vanholder R, et al., Toxins 2018, 10, 33. https://doi.org/10.3390/toxins10010033 [2] Masereeuw R, et al., Semin Nephrol. 2014 Mar;34(2):191-208. [3] Patent US2014246367(A1): Jankowski J, et al., Priority year 2011. [4] Patent WO2014095072(A1): Tschulena U, et al., Priority year 2012. [5] Patent EA201500722(A1): Jankowski J, et al., Priority year 2013. [6] Van Gelder et al, Biology 2021, 10, 292. https://doi.org/10.3390/biology10040292 16:00 0 mins The role of wrist-worn technology in telemonitoring Parkinson’s disease in daily life: a narrative review Peng Li, Richard van Wezel, Ying Wang Abstract: Introduction: Parkinson’s disease (PD) is a worldwide fast-growing neurological disorder mainly affecting the 65 and older. Healthcare professionals can only diagnose PD when patients present motor symptoms at the late disease stage according to the UK Brain Bank Criteria [1]. The individual’s disease progression is small and heterogeneous in the daily living environment which makes timely diagnosis difficult. Therefore, it is very important to track the small changes along with the disease progression during daily life to take effective interventions to prevent the disability. Telemonitoring systems with wearable sensing technology can assist healthcare professionals to gain insights into patients' daily life health conditions and help patients empower themselves to improve treatment effectiveness and mitigate disease progression [2-3]. Wrist-worn digital devices, especially smartwatches, are currently the most popular tool in the PD research field given their convenient usage during long-term daily life monitoring. Even though the telemonitoring technology of PD using wrist-worn sensing has attracted great interest with numerous articles, there is still unclarity about the value of using this technology in PD clinical practice. In this narrative review, we evaluated the challenges and opportunities of wrist-based technology in the early diagnosis, progression tracking, and self-management of PD. Methods: Peer-reviewed articles were searched in October 2022 from four databases: PubMed, Web of Science, IEEE Xplore, and Google Scholar. The search keywords are (“Parkinson” OR “PD” OR “Parkinson’s disease”) And (“wrist” OR “wrist-worn” OR “Smartwatch” OR “Fitness band”). We selected articles and extracted useful information from the articles, such as the sensors and algorithms that have been used, the features of monitoring, the monitored motor or non-motor symptoms in PD, applications in real clinical practice (early diagnosis, progression, disease management), the comfortability and usability during long time usage, etc. Results & Conclusion Further results and conclusions will be presented in the near future based on the findings from the literature review. Reference: 1. Calne DB, Snow BJ, Lee C. Criteria for diagnosing Parkinson’s disease. Ann Neurol (1992) 32:S125–7. 10.1002/ana.410320721 2. Dorsey ER, et al. “Moving Parkinson care to the home”. Movement Disorders. 2016 Sep;31(9):1258-1262. DOI: 10.1002/mds.26744. 3. Dorsey, E Ray, and Eric J Topol. “Telemedicine 2020 and the next decade.” Lancet (London, England) vol. 395,10227 (2020): 859. doi:10.1016/S0140-6736(20)30424-4 * Corresponding author: Peng Li, pli-1@utwente.nl 16:00 0 mins Using 2D CNN to detect tonic-clonic seizures based on accelerometer and photoplethysmography signals Chunjiao Dong, Johannes van Dijk, Xi Long, Ronald M. Aarts Abstract: This study aims to design a deep learning model to automatically detect tonic-clonic (TC) seizure events based on accelerometer (ACM) and photoplethysmography (PPG) signals, which are vital signals to illustrate the motion and heart rate changes during TC seizures [1]. Both signals were continuously collected using NightWatch armbands [2], from 44 patients during the night, and each patient was monitored for two to three months. 16:00 0 mins Ultrasound-based automatic 3D modulography for atherosclerotic carotid artery plaques Koen Franse, Jan-Willem Muller, Marc van Sambeek, Richard Lopata Abstract: 1. Introduction Currently invasive surgery on atherosclerotic plaques in the carotid artery is performed in case of stenosis >70% [1]. However, retrospective studies show that this intervention is only needed in one out of six patients. Mechanical modelling of the arterial wall with patient-specific local material parameters can improve clinical decision making by characterizing the plaque’s morphology and prevent rupture. This study presents a 3D inverse finite element (FE) algorithm that can locally reconstruct the heterogeneous characterization of the arterial plaque based on non-invasive ultrasound (US). 2. Methods The input for the inverse FE algorithm is strain data, which can be obtained by performing strain imaging on US data. In this study, cross-sectional 2D three-angle plane wave US acquisitions are simulated using the k-wave toolbox [2]. The simulations are based on a 3D FE model of a plaque containing a lipid pool and calcifications. The inverse FE workflow initiates with a homogeneous plaque model. The ratio between the FE stress and the input strain is used to compute the apparent Young’s modulus (AYM) locally [3]. This parameter is further optimized to minimize the error between the input strain and the iteratively computed FE strain. The resulting material parameters are compared with those in the ground-truth FE model for two cases: the inverse reconstruction based on FE strain input data and the reconstruction based on US simulation strain imaging data. 3. Results & Discussion Using FE strain data as input, the 3D inverse algorithm demonstrates the ability to reconstruct the 3D geometry of the plaque, including details of the lipid pool and small calcifications, without prior segmentation. However, using the 2D plane wave US strain imaging data as input, the algorithm is not able to reconstruct local plaque details. The poor performance of the US strain-based reconstruction is due mainly to the lack of precision and resolution in the lateral US strain data. Further research will be aimed at improving this by using more compounded plane wave angles and FE-based regularization methods. [1]: Leo H Bonati et al., “European Stroke Organization guideline on endarterectomy and stenting for carotid artery stenosis”, European Stroke Journal, Vol. 6, No. 2, I-XLVII (2021) [2]: B. E. Treeby and B. T. Cox, “k-Wave: MATLAB toolbox for the simulation and reconstruction of photoacoustic wave fields”, J. Biomed. Opt. Vol. 15 No. 2, 021314 (2010) [3]: J. Porée, B et al., “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, 1805-1817 (2017) 16:00 0 mins Adhesive for fixation of polycarbonate urethane implants to the bone Pardis Farjam, Edsko Hekman, Gijsbetus Verkerke, Jeroen Rouwkema Abstract: Polyurethanes (PU) have been used in orthopaedic implants due to their superior characteristics including; good biocompatibility and mechanical properties, lubricity, and good wear properties. PU can be involved in a diverse range of orthopaedic applications including but not limited to joint replacement implants. We are currently designing a novel PU-based joint replacement prosthesis, for which we are investigating adhesives as a fixation mechanism to the bone. In orthopaedics, bone adhesives are commonly designed to treat simple and comminuted fractures. Bone adhesives could also be employed as the fixation technique of implants to bone. Adhesives as the anchorage tool bring several advantages, such as: preserving the integrity of the tissue and the implant with the possibility to be delivered via minimal-invasive techniques and offering simple and precise applicability. To be employed as an implant fixation technique, sufficient strength of an adhesive is one of the dominant requirements. Cyanoacrylate-based adhesives have shear bond strengths in the range of 1-2 MPa in a bone-bone bond. Purpose In this study, we evaluated a commercial biocompatible cyanoacrylate-based adhesive Glubran2® (Gem, Italy) as a candidate fixation technique for PU-based orthopaedic implants. Materials and Method PU film was obtained from our project partners at the Fraunhofer Institute for Manufacturing Engineering and Automation (Stuttgart, Germany). Square specimens of 25 mm*25 mm with a bonding area of 25 mm*10 mm were used. A lap-shear test has been conducted according to ASTM standards F2255 for strength properties of tissue adhesives in lap-shear by tension loading. Results The apparent shear strength as the maximum load divided by the bond area was revealed to be 0.07± 0.01 (MPa). Failure did predominantly occur at the bonding site between the cyanoacrylate-based adhesive and the PU film. Conclusion The biocompatible cyanoacrylate-based adhesive showed inferior shear strength in bonding PU to bone compared to bonding bone to bone. New fixation candidates will be studied for the fixation of PU-based joint replacement prostheses. Acknowledgements This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 863183. 16:00 0 mins Augmentation of medical images with unsymmetrical intensity distribution Rajarajeswari Ganesan, Antonino Amedeo La Mattina, Frans van de Vosse, WOUTER HUBERTS Abstract: Background: In Silico Clinical Trials (ISCT) display promising advances over Human Clinical Trials for the assessment of safety, efficacy and usability of new diagnostic/interventional procedures or medical devices. One of the reasons for this is ISCTs allow for data augmentation by scaling up both the number of patients and different patient phenotypes present in the population. In our work, we aim for augmentation of Hip Fracture Scans using deep generative model. Method: We propose a novel Generative Adversarial Networks (GANs) architecture. In GANs, the Generator aims to generate images like real images and the Discriminator labels them as realistic/unrealistic. In this architecture, we have introduced Neural Conditional Random Fields (NCRF) within the Deep Convolutional Network(DCN) to capture the non-symmetrical intensity relations present in Hip Fracture CT Scans. These non-symmetric intensity variations distribute the entire image into different regions of the femur. The sharp variations determine the fine structure. In addition, there is a relation between the neighbouring intensity and the fine structure. In the DCN, the adversarial pair learns hierarchy of representations from main edges to fine structures in an unsupervised manner. In our method, the generator with NCRF contemplates the spatial relations within the neighbouring patches. This NCRF layer is directly integrated with each convolutional layer. This consolidated feature map of each layer is fed as an input for the next layer. This architecture helps the network learn the spatial relation at every level. Thereby, the generator learns the features along with the spatial relation. This approach helps us to retain the intensity relation. Results: Though the fine structure in the images are still indistinguishable, the quantitative metrics such as Fréchet Inception Distance (FID), Structural Similarity Index (SSIM) and Peak Signal-to-Noise Ratio (PSNR) match the predefined standard for the generated images. Conclusion: The network captures the spatial relation but acute structures generated by the network are unclear. Therefore, we will work towards improvement of the network for more realistic results 16:00 0 mins Guided sacroiliac joint fusion by virtual surgical planning superimposed on intraoperative fluoroscopic images using digitally reconstructed radiographs steven Lankheet, Nick Kampkuiper, Edsko Hekman, Jorm Nellensteijn, Maaike Koenrades, Femke Schröder Abstract: Sacroiliac joint (SIJ) fusion is a surgical treatment for SIJ dysfunction. During surgery, implants are placed through the SIJ under fluoroscopic guidance using a minimally invasive operation technique. The procedure aims to stabilise the joint and thereby reduce pain. While placing the implants freehandedly, closely located nerve structures must be avoided to prevent complications. However, these adjacent structures are not clearly distinguishable under fluoroscopic guidance. Besides, the anatomy of the sacrum is extremely variable. This challenges safe and effective implant placement in a stable configuration. Preoperative virtual surgical planning (VSP) may help the surgeon to achieve optimal implant positioning and lower the risk of complications, operation time and radiation dose. However, to accurately recreate the VSP in the operating room is challenging since the VSP cannot be directly compared to intraoperative fluoroscopic images. Therefore, this study aimed to develop and test a method that superimposes a VSP onto intraoperative fluoroscopic images using digitally reconstructed radiographs (DRRs). To this end, an algorithm was developed using ray casting, registration and visualisation techniques. The algorithm was tested during a routine SIJ fusion surgery. The workflow is as follows. The preoperative CT scan and VSP are loaded into the software to prepare a series of low-resolution DRRs under different angles. During surgery, a lateral fluoroscopic image is loaded into the software and the optically best matching DRR is determined. This DRR is used to create a high-resolution image with overlaid VSP, showing the position of guide pins used for implant placement. The surgeon then places the pins in the same position and angle as represented on the DRR to create a comparable fluoroscopic image. The developed intraoperative workflow was found to be fast and feasible for clinical use. A postoperative placement accuracy analysis showed a mean 3D deviation at the apex of the implant of 4.6 mm, a mean angle deviation of 3.6º and a mean entry point deviation of 3.4 mm. Further research is needed to optimise the developed workflow and clinically validate the proposed method. 16:00 0 mins Estimating Haemodialysis adequacy using urea monitoring in sweat or saliva Sophie Adelaars, Stijn Konings, Lieke Cox, Massimo Mischi, Arthur Bouwman, Daan van de Kerkhof Abstract: In a large proportion of patients with end-stage renal failure, haemodialysis (HD) adequacy is performed multiple times a week, which results in high disease burden. To determine clearance adequacy in HD, the urea concentration is analysed in blood samples taken before and after HD cycles. To enable HD to be performed in a home situation, which is a promising future development to improve patient quality of life, there is a clinical need to perform continuous and non-invasive monitoring of urea kinetics without taking blood samples. To this end, saliva and sweat are potential alternative matrices to monitor HD adequacy. The aim of this study was to determine whether the urea concentration in sweat and saliva is associated with the plasma concentration in HD patients. Sweat and salivary urea concentrations were analysed at the start and at the end of one HD cycle and compared to the corresponding plasma concentrations. Here, we report interim results (N = 31) in a study that is aimed to include 40 HD patients in total, which started October 2021. Concentrations in sweat were significantly higher than plasma concentrations at the start and end of HD (both p<0.001). Salivary concentrations were also significantly higher at the start (p<0.01), but not at the end of HD. In plasma, the urea concentration decreased from 21.50 (17.00 – 25.25) mmol/L to 5.75 (4.15 – 7.25) mmol/L (median (IQR)) in the studied HD cycle. A decrease was also observed in sweat, from 27.30 (21.70 – 32.78) mmol/L to 11.01 (7.87 – 14.42) mmol/L, and in saliva, from 24.60 (21.05 – 29.55) mmol/L to 5.63 (4.02 – 7.11) mmol/L. Urea concentrations in plasma showed a significant correlation (p<0.001) with sweat urea concentrations (Spearman’s correlation r 0.932 (0.873 – 0.964)) and with salivary urea concentrations (r 0.929 (0.885 – 0.957)). Although the findings of this interim analysis are preliminary, the results illustrate a proof of principle of urea measurements in sweat and saliva to monitor HD in a non-invasive continuous manner. These findings indicate that studies aimed at developing data algorithms and bio-sensing methods using sweat or saliva for application in a clinical setting deserve further attention. 16:00 0 mins Optical coherence tomography in the femoropopliteal track Lisa Rutten, Lennart van de Velde, Michiel van Werkum, Kartik Jain, Michel Versluis, Michel Reijnen Abstract: Objective Peripheral arterial disease (PAD) is one of the most common vascular diseases that limits a patient’s walking capability and can in severe cases lead to amputations.[1] Treatment of PAD often requires stent placement.[2] Stent sizing and positioning influence the stent patency.[3] Current treatment planning, based on digital subtraction angiography (DSA), is limited as DSA only visualizes the vessel lumen in 2D, possibly resulting in suboptimal stent sizing and positioning. This can lead to unfavourable wall shear stress levels, causing restenosis and stent failure.[3] Treatment planning can be improved by intravascular optical coherence tomography (OCT) as OCT visualizes the vessel lumen, vessel wall and stent in 3D with micrometer resolution. In this study, DSA-based and OCT-based treatment planning will be compared to investigate the added value of OCT in the treatment of PAD. Furthermore, technical success of the OCT measurements will be assessed. Methods 25 patients diagnosed with PAD and scheduled for endovascular treatment with the Supera stent will be included. Pre and post stent placement, DSA scans followed by OCT scans will be performed. Directly after each scan, the findings and the treatment plan will be noted. For each OCT scan technical success, defined as complete vessel in field of view and blood clearance, will be reported. Results To date, five patients have been enrolled. OCT changed the DSA-based treatment plan 4 times, because of the presence or absence of a stenosis (n=2), a dissection (n=1) and stent strut malapposition (n=1), that were all not observed with DSA. Changed treatments included choosing a longer stent (n=2), performing post dilation of the stent (n=1) and not performing balloon angioplasty (n=1). Full and partial vessel visibility was achieved in 70% and 26% of the OCT scans, respectively, and complete blood clearance in 70%. Conclusions These preliminary results show that OCT can detect new findings that change the endovascular treatment plan of PAD. Furthermore, technical success was achieved in the majority of the OCT scans. Inclusion is still ongoing and the obtained OCT scans will be used to develop reliable computational fluid dynamics models that could predict stent patency. 16:00 0 mins Combining mechanistic modeling and machine learning to model slow hemodynamic changes in ICU patients Roy van Mierlo, Natal van Riel Abstract: Due to the limited time available in intensive care units (ICUs), physicians often have difficulty deriving an unstable patient’s underlying pathology. Regularly, reactive treatments are started for these patients, whereas timely functional examination can give more insight into the underlying pathology of instability. Therefore, predicting instability would allow for proactive and more specialized interventions and could even prevent deterioration in risk patients altogether. Previous attempts at data-based prediction of hemodynamic instability have been made.[1] However, incorporating physical laws can allow for more accurate instability prediction and, in addition, can help with determining underlying pathologies and their correct treatment strategies. Arterial wall mechanics and arterial blood pressure directly characterize the functionality of the cardiovascular system and thus give clear insights into a patient's hemodynamic status.[2] Although measuring such quantities in vivo gives highly accurate results, placing additional probes in patients in life-threatening situations is not ideal. Deriving these quantities in a non-invasive manner has motivated the use of computational models for in silico prediction. The goal of this study is to leverage hybrid modeling techniques (combining mechanistic modeling and machine learning) to model hemodynamic changes over a patient’s ICU stay. We use an arterial blood pressure waveform over a short period (seconds) and consider a two-stage Windkessel model to derive the model parameters (resistances and compliances for the large and small vessels). Next, by deriving these parameters at several stages of a patient’s ICU stay, we can find a model for the long-term behavior (days) of these Windkessel parameters using machine learning.[3] Patient-specific model parameters of this long-term behavior model will have good prognostic value for predicting instability and certain hemodynamic complications. References 1. Rahman, A., Chang, Y., Dong, J., Conroy, B., Natarajan, A., Kinoshita, T., ... & Xu-Wilson, M. (2021). Early prediction of hemodynamic interventions in the intensive care unit using machine learning. Critical Care, 25(1), 1-9. 2. O’Rourke, M. (1995). Mechanical principles in arterial disease. Hypertension, 26(1), 2-9. 3. Regazzoni, F., Chapelle, D., & Moireau, P. (2021). Combining data assimilation and machine learning to build data‐driven models for unknown long time dynamics—Applications in cardiovascular modeling. International Journal for Numerical Methods in Biomedical Engineering, 37(7), e3471. 16:00 0 mins Towards a digital twin of a growing fetal heart Bettine van Willigen, Beatrijs van der Hout-van der Jagt, Wouter Huberts, Frans van de Vosse Abstract: The Perinatal Life Support (PLS) consortium is developing a liquid-based environment (PLS system) to increase chance of survival of extremely preterm infants (< 28 weeks of gestational age). To develop such a complex device, knowledge from multidisciplinary fields must integrate into one single system. Mathematical models are used to support this integration by composing a digital twin of the PLS system and the preterm infant to allow computer simulations of the interaction with the device. The digital twin consists of a closed-loop fetal cardiovascular 0D circulation model connected to a 1D wave propagation model that describes the larger vessels. Furthermore, the contractile behaviour of the four cardiac chambers are included by a one fiber model1 for each of them. This model describes the underlying phenomena of cardiac contraction by relating cavity pressure to fiber stress, cavity and wall volume, and the geometrical inputs of this model makes it possible to simulate fetal cardiac growth without the need to redefine the contraction model. However, the mathematical models that describe the valve leaflet motion for 0D and 1D models2–4 seem not sufficient to allow fetal cardiac growth without fluttering of the valve leaflets. Part of the problem lies in the non-uniqueness of the parameter set of these models as these are chosen and tuned for a specific case. Therefore, these models do not always result in physiological realistic outcomes when growth of the cardiac valves are initiated based on scaling laws or when clinical data is used for its digital twin purpose. In this study a general formula describing three different 0D valve models at once has been developed: the ideal diode, the valve model of Mynard et al.2, and the valve model of Korakianitis and Shi4. These three models have all three a different assumption about the relation between the pressure drop over and flow through the valve. Furthermore, these valves describe motion of the valve leaflets differently. By performing a sensitivity analysis5 on the general description of these valves, the importance of the parameters and the relation between the parameters can be identified. Furthermore, the correctness of the assumptions can be analysed. Ultimately, a valve model can be defined that describes the underlying phenomena of fetal cardiac valves in order to allow digital twinning and, thus, fetal cardiac growth. 16:00 0 mins The effect of thickness and stiffness of a titanium inlay in a cementless peek femoral component on the primary fixation and bone strain energy density Corine Post, Thom Bitter, Inger van Langen, Adam Briscoe, Nico Verdonschot, Dennis Janssen Abstract: Introduction: Polyetheretherketone (PEEK-OPTIMATM) is used as an alternative material for total knee arthroplasty (TKA) components as its stiffness characteristics more closely resemble the stiffness characteristics of human bone compared to the currently used metal alloys. This on the long-term may avoid stress-shielding of the TKA. Earlier research on a cementless PEEK femoral component suggested that larger bone-implant interface micromotions were found relative to a cementless cobalt-chrome femoral component. Porous titanium has been widely used for facilitating bone ingrowth and therefore, a titanium inlay on the inside of a cementless PEEK femoral component may reduce the micromotions as well as preserving the beneficial (low) stiffness characteristics. Therefore, the objective of this study was to assess the effect of varying the thickness and stiffness of a titanium inlay on the inner surface of a cementless PEEK femoral component on the primary fixation and bone stress-shielding using finite element (FE) analysis. Methods: FE models of the femur and femoral component were created with five variants of titanium inlays: thin, medium, thick, distal thick – proximal thin, and distal thin – proximal thick. The axial forces of a jogging activity from the Orthoload database were applied on the medial and lateral femoral condyles. The 95th percentile of the maximum resulting micromotions and the strain energy density (SED) were quantified as outcome measures. Results: The addition of a titanium inlay reduced the micromotions. The largest micromotions arose on the medial side of the anterior flange. The distal thick – proximal thin variant showed the lowest micromotions. The SED mainly decreased in the distal region. This was more pronounced for a stiff and thick inlay. Conclusion: This study showed that the combined distal thick – proximal thin inlay gives the best trade-off between reducing micromotions and prevention of periprosthetic stress-shielding. The addition of a titanium inlay on the inner surface of a cementless PEEK femoral component may therefore result in a longer lifespan of the femoral component. Acknowledgements: PEEK-OPTIMATM is a trademark of Invibio Ltd. Implant geometry was supplied by Maxx Orthopaedics Inc. 16:00 0 mins Quantitative assessment of carotid diameter measurements in parallel versus rotated and tilted orientation using ultrasound in the operating room – a comparative analysis Esmée de Boer, Catarina Dinis Fernandes, Danihel van Neerven, Christoph Pennings, Rohan Joshi, Sabina Manzari, Sergei Shulepov, Luuk van Knippenberg, John van Rooij, 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 cardiac output estimation. 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, obtaining and maintaining this mid-axis parallel view is difficult, and 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. Preliminary research on velocity measurements showed that the rotated and tilted view was more robust to motion and less operator-dependent than the parallel view. To evaluate these findings in a clinical setting, we performed measurements with a parallel and rotated and tilted probe orientation intending to investigate the operator-dependency and diameter estimates. Transverse measurements were also obtained and used as a reference. Carotid Doppler measurements of 25 adult cardiac surgery patients were performed with the probe oriented parallel, transverse, and rotated and tilted. The US recordings were analyzed with previously developed algorithms, optimized for each probe orientation, resulting in a median diameter estimation per 30s measurement. Preliminary results of the first 9 patients show that diameter estimations from the three views are comparable: median diameter estimates [IQR] are 6.27 [5.67-7.25]mm, 6.74 [6.33-7.13]mm, and 6.33 [5.98-7.69]mm for the parallel, transverse, and rotated and tilted views, respectively. Visual data inspection confirmed comparable within-patient estimates in all except two patients, in whom artifacts caused by probe movement were found in the US recordings. Normality was checked using the Shapiro-Wilks test and a one-way ANOVA confirmed that the diameter estimates from the different views were not significantly different. Future analysis comprises expansion of the current results and investigating inter- and intra-operator reliability of the different views. 16:00 0 mins Radiation safety for clinicians in the cathererization laboratory Thijs van Deudekom Abstract: Introduction and objectives - Radiation dose during cardiac catheterisation interventions has been a topic of growing interest in interventional cardiology in recent years. As it has adverse impacts on both the patient and the medical staff, there is great interest in decreasing radiation exposure during these procedures. For instance, interventional cardiologists have a higher chance of developing cataracts, lens opacities, skin cancer and brain tumours. Radiation protection consists of active and passive components. The active component focuses on minimising radiation during surgery. Active protection strategies include routine and appropriate use of lead apparel, using techniques in reducing radiation use to the patient and thereby the operator, beam angulation and repositioning yourself in the room. Unfortunately, clinicians do not realise when they receive too much radiation and can deploy certain measures. In current care, there is often monthly feedback on how much radiation the person has received, instead of during or directly after procedures. Real-time dose monitoring is upcoming, however, several types of feedback are being offered to staff. In this review the effectiveness of different possible real-time feedback methods are compared. Results - Possible feedback mechanisms include sounds, a bracelet displaying radiation, a 2D computer graphic of the patient and table from the ceiling viewpoint, and a 3D augmented reality (AR) display with a real-time video feed of the intervention room. However, the most commonly used and most promising technique was a bar graph displayed on a screen for the operator to see. The current dose rate was displayed in colour bars, which increased in size and changed colour as the radiation thresholds changed. Discussion - Audible feedback also resulted in a partial radiation reduction. However, its standard use was not tolerated by the staff and appeared to be too distracting. Further development is necessary to provide better feedback with more specific instructions on how to optimally use the possible measure. For instance, how to place the lead screen or to increase the distance to the source. Ultimately, automation or robotisation would be the optimal solution, as the doctor could be at a distance and would receive minimal radiation. 16:00 0 mins Development of a perfusion phantom for cerebral perfusion imaging Liselot Goris, Marije Kamphuis, Abdallah Zaid Al-Kaylani, Henny Kuipers, Johan van Hespen, Richte Schuurman, Reinoud Bokkers, Srirang Manohar Abstract: Purpose Time from symptom onset to treatment is one of the most important modifiable factors affecting functional outcome in patients with acute ischemic stroke.1 Computed tomography perfusion (CTP) imaging is an integral part of the standard diagnostic workup. Using cone-beam computed tomography (CBCT) perfusion imaging in a direct-to-angio approach might significantly reduce workflow times and improve outcome.2 However, the diagnostic accuracy of CBCT perfusion must first be comparable to CTP.2 There is a scarcity of real-patient data regarding the diagnostic accuracy of CBCT perfusion in acute ischemic stroke.2 This study aims to develop a cerebral perfusion phantom setup to facilitate ground-truth evaluation of CTP, and to compare between a validated CTP standard and CBCT. Methods The phantom is designed for 3-D printing and comprises a vessel structure that branches into two microcirculation compartments. The compartments can be filled with a material of choice which mimics the flow dynamics of the contrast. The flow can be measured with flow sensors and controlled with three-way valves to simulate a perfusion defect. As an intermediate step, the reproducibility is evaluated by acquiring 2-D perfusion images with a C-arm system and analysing the differences in the time-intensity curve (TIC) parameters. Proof-of-concept tests are performed with CTP and its associated clinical software to test whether the phantom can achieve realistic HU densities, arterial input function (AIF) and microcirculation similar to those found in normal human brain tissue. Results The maximum standard deviation of TIC-parameters between repeated measurements is 4.4%. In CTP measurements the AIF's TTP range is 5-10 seconds. Baseline-to-peak intensity is 250 HU. The TIC of the microcirculation has a TTP of 20 seconds and baseline-to-peak intensity of 60 HU. CBF ranges from 155-440 mL/100gr tissue/min. The computed and measured flow were directly related (ρ = -0.98, p=0.01). Discussion and Conclusions Future work is needed to tune the setup to obtain CBF values more similar to human cerebral perfusion, 10, 20 and 60 mL/100gr tissue/min in ischemic core, penumbra and healthy tissue respectively. Thereafter, a validated CTP standard will be defined, and the setup will be used to evaluate perfusion imaging in CBCT. 16:00 0 mins Local assessment of mechanical failure and structural properties of human carotid fibrous plaques Su Guvenir Torun, Pablo de Miguel Munoz, Hanneke Crielaard, Hence Verhagen, Gert-Jan Kremers, Antonius van der Steen, Ali C. Akyildiz Abstract: Biomechanical analysis of atherosclerotic plaque rupture is crucial since it is the local failure of fibrous plaque tissue. Currently, the knowledge on mechanical failure properties of fibrous plaque is scarce [1,2], and limited to global properties based on homogeneity assumption. However, the fibrous plaque has a highly collagenous heterogeneous structure, and its failure properties should be investigated locally, together with the underlying collagen structure information. Yet, only one study [1] investigated the structural and failure properties of carotid fibrous plaques, but it was limited to the global failure properties obtained at the ultimate failure. However, initiation of fibrous plaque rupture occurs before the ultimate failure. In this study, for the first time, we investigated local failure properties and the collagen structure of human carotid fibrous plaques at the rupture initiation. The pipeline consists of uniaxial tensile testing combined with digital image correlation, for obtaining local deformation at the failure, and second harmonic generation imaging, for the assessment of local collagen predominant angle (pDA) and dispersion of fibrous plaque strips (n=30). In structural analysis, the pre-testing collagen pDA in ~90% of measured local regions of all strips were observed to have close to circumferential orientation, with a varying level of local dispersion. Similarly, rupture regions were also observed to have close to circumferential pDAs. In mechanical analysis, the comparison of global strain measurements to local DIC measurements showed that the global analysis greatly underestimated tissue rupture strain. In addition, the locations of the high circumferential strain regions were correlated with the rupture regions. To the best of our knowledge, this study is the first to characterize local mechanical failure and structural collagen of fibrous plaques at rupture initiation. Findings from this work highlight the importance of patient-specific local assessment of mechanical failure and structural properties of the highly heterogenous fibrous plaques. The circumferential strain appears to be an important candidate for local rupture initiation assessment, whereas the structural collagen parameters alone were not observed to be sufficient. For future work, mechanical testing and structural imaging could be combined for the assessment of collagen re-alignment and local strains at rupture initiation. [1] Johnston R et al., Acta Biomaterialia, 124: 291-300, 2021. [2] Teng Z et al., Journal of Biomechanics, 48.14: 3859-3867, 2015. 16:00 0 mins Ultrasound blood velocity imaging in hard-to-image carotid arteries using cascaded dual-polarity waves Joosje de Bakker, Chris de Korte, Anne Saris Abstract: Stroke is a leading cause of death and major disability and is often caused by atherosclerosis. Complex blood flows (CBF) in the carotid arteries seem to play a crucial role in the atherosclerotic disease process. However, conventional ultrasound systems cannot measure these CBF. Ultrafast ultrasound, where thousands of images are acquired per second, enabled new opportunities for blood velocity imaging. Identification of CBF could aid in better patient-specific risk stratification and treatment planning. A downside of ultrafast ultrasound, where unfocused beams are transmitted, is the decreased signal-to-noise ratio (SNR) compared to conventional focused imaging. Cascaded Dual-polarity Waves (CDW)a overcomes this by transmitting a pulse-train instead of a single pulse. A postprocessing decoding scheme is applied to obtain recovered signals with increased SNR. This potentially enables blood velocity imaging under challenging conditions, such as deeper located vessels. However, its application in blood velocity estimation has not been shown yet. The performance of 4-CDW was compared to single pulse plane waves (PW) using Field IIb,c simulations of parabolic flow (vmax = 0.25, 2 m/s). Different attenuation situations were mimicked by adding two different noise levels (SNR = 6, 14 dB). Data were simulated for a 7.8 MHz linear array, subsequently transmitting -20° and +20° beams at a repetition frequency of 8 kHz. Blood velocities were estimated using normalized cross-correlation based compound speckle trackingd. For the peak velocity of 0.25 m/s and 14 dB SNR, CDW and PW performed very similar. For a reduced SNR of 6 dB, CDW demonstrates a thirteen fold lower lateral bias than PW. Increasing vmax to 2 m/s raised the bias for both imaging methods. PW outperformed CDW in the 14 dB SNR case, this reversed in the 6 dB case. Summarized, in low SNR conditions, CDW consistently outperformed PW due to the increased effective SNR for the recovered signals. However, since high velocities result in imperfect summation/cancellation of shifted pulse responses, CDW fails to outperform PW in case of sufficient SNR. To conclude, CDW could potentially aid in measuring CBF in high attenuating situations, such as in obese patients, enabling blood velocity imaging in all patients. 16:00 0 mins AI Aerobic Exercise coaching for seniors: Improving cardiovascular fitness on Aruba Francis Laclé, Sultan Salys, Bart Vanrumste Abstract: Aruba, an island in the Caribbean and part of Kingdom of the Netherlands, has seen in the past decades an upwards trend in life expectancy that is comparable to the regional average [1]. Yet, mortality rates attributed to circulatory causes (including cardiovascular) reduce the life expectancy to 1.02 and 0.56 years for male and female inhabitants respectively, when compared to the Netherlands [2,3]. Cumulative trend analyses on mortality data for seniors (65+) shows cardiovascular diseases (CVDs) to be the top five for Aruba [2]. Fortunately, research has established that exercise and physical activity do decrease the risk of CVD [4–7]. To help reduce CVD incidence, we aim to investigate solutions that innovate on movement sensing, tracking, and coaching technologies. From an engineering perspective, the technical challenge is to overcome gaps in the real time modelling of cardiorespiratory fitness (CRF) from affordable hardware, and coupling CRF optimization with reinforcement learning. Our main research question is as follows: given zero human intervention, can an agent coach its users to improve their CRF through aerobics exercises within a predetermined period? Two sub-questions are how to accurately model CRF for future aerobics coaching, and second, how to reliably improve CRF with reinforcement learning agents and coaching feedback mechanisms. First, we will conduct cohort studies with local participants to collect data to address the first sub-question. These cohort studies consist of three phases, with maximal tests (measuring VO2max) carried during the first and third phase. During the second phase, we will measure intensity variables such as heart rate, 3D body and joint accelerations, and instantaneous VO2, while participants follow a fixed subset of aerobic exercise routines. By adopting a multimodal approach to measuring intensity, we increase our search space to find a reliable CRF estimator for a fixed subset of aerobic exercises. To increase the financial viability of implementing a new coaching technology that is applicable in small island states such as Aruba, we also prioritize sustainment by considering tradeoffs between innovation and the cost of soft- and hardware procurement and maintenance. REFERENCES 1. United Nations, Department of Economic and Social Affairs (2022) World Population Prospects: The 2022 Revision: Data Portal. https://population.un.org/dataportal/data/indicators/61/locations/533,900,904/start/1950/end/2022/line/linetimeplot Accessed 21 Dec 2022. 2. Verstraeten SPA, van Oers HAM, Mackenbach JP (2020) Contribution of amenable mortality to life expectancy differences between the Dutch Caribbean islands of Aruba and Curaçao and the Netherlands. Rev Panam Salud Publica 44: e38. http://www.ncbi.nlm.nih.gov/pubmed/32435265 3. Central Bureau of Statistics (2022) Gender Statistics: Health and related services. https://cbs.aw/wp/index.php/2019/11/18/gender-statistics/#health Accessed 21 Dec 2022. 4. Blair SN (1989) Physical Fitness and All-Cause Mortality: A Prospective Study of Healthy Men and Women. JAMA 262: 2395. https://doi.org/10.1001/jama.1989.03430170057028 5. Thompson PD, Buchner D, Pina IL, Balady GJ, Williams MA, et al. (2003) Exercise and physical activity in the prevention and treatment of atherosclerotic cardiovascular disease: a statement from the Council on Clinical Cardiology (Subcommittee on Exercise, Rehabilitation, and Prevention) and the Council on Nutrition, Physical Activity, and Metabolism (Subcommittee on Physical Activity). Circulation 107: 3109–3116. http://www.ncbi.nlm.nih.gov/pubmed/12821592 6. Wilmot EG, Edwardson CL, Achana FA, Davies MJ, Gorely T, et al. (2012) Sedentary time in adults and the association with diabetes, cardiovascular disease and death: systematic review and meta-analysis. Diabetologia 55: 2895–2905. http://www.ncbi.nlm.nih.gov/pubmed/22890825 7. Ross R, Blair SN, Arena R, Church TS, Després J-P, et al. (2016) Importance of Assessing Cardiorespiratory Fitness in Clinical Practice: A Case for Fitness as a Clinical Vital Sign: A Scientific Statement From the American Heart Association. Circulation 134: e653-e699. http://www.ncbi.nlm.nih.gov/pubmed/27881567 16:00 0 mins Low-cost-shaping of electrical stimulation waveforms for bioelectronics medicine with improved efficiency and selectivity Amin Rashidi, Francesc Varkevisser, Tiago Costa, Vasiliki Giagka, Wouter Serdijn Abstract: Electrical stimulation is proven to be an effective way of neuromodulation in bioelectronic medicine (e.g. cochlear implants, deep brain stimulators, etc.), delivering localized treatment by the means of electrical pulses. To increase the stimulation efficiency and neural-type selectivity, there is an increasing interest to employ non-rectangular stimulation waveforms [1-4]. Even though delivering and storing digital data at the stimulator provides the highest flexibility for generating stimulation waveforms, state-of-the-art approaches suffer either from poor resolution or the requirement of high data bandwidth for wirelessly powered implants [2]. Using Analog waveform generators is an alternative approach at the cost of extra implementation complexity for each type of waveform [3]. To fulfill the same goals as employing arbitrary waveforms for stimulation, we propose to shape the typical rectangular waveform using a programmable first-order low-pass filter, mimicking the natural filtering characteristic of the neural membrane. Using bio-realistic modeling, we show that such a pre-filtered waveform requires less or equal energy for the activation of neurons when compared with other energy-efficient waveforms (e.g. Gaussian). Notably, this comes at the low cost of only one extra programmable parameter (i.e., the filter’s corner frequency), on top of the typical duration and amplitude parameters. The basic concept of this work is driven by the fact that the natural low-pass characteristic of the neuron’s membrane limits the energy transfer efficiency from the stimulator to the cell. Thus, it is proposed to pre-filter the high-frequency components of the stimulus [4]. The method is validated for a Hodgkin-Huxley (HH) axon-cable model using NEURON v8.0 software. The required activation energy is simulated for rectangular, Gaussian, half-sine, triangular, ramp-up, and ramp-down waveforms, all with pulse durations of 10-1000µs, and low-pass filtered with cut-off frequencies of 0.5-50kHz. Simulations show a 51.5% reduction in the required activation energy for the shortest rectangular pulse (i.e., 10-μs pulse width) after filtering at 5kHz. It is also shown that the minimum required activation energy can be decreased by 11.04%, 9.49%, 8.28%, 1.81%, 0.17%, and 0% when an appropriate pre-filter is applied to the rectangular, ramp-down, ramp-up, half-sine, triangular, and Gaussian waveforms, respectively. Finally, a perspective usage of this method to improve the selectivity of electrical stimulation is drawn. 16:00 0 mins High framerate ultrasound for the analysis of local blood flow parameters in relation to limb occlusion after EVAR, a patient specific in vitro study Hadi Mirgolbabaee, Guillaume Lajoinie, Bob Geelkerken, Michel Versluis, Erik Groot Jebbink, Michel Reijnen Abstract: Background: Endovascular aneurysm repair (EVAR) has become the primary treatment option for abdominal aortic aneurysms (AAA). Although EVAR has excellent short and mid-term outcomes concerning mortality rates and hospitalization, its long-term results showed a similar survival rate as open surgical repair1,2. Moreover, 20-30% of EVAR-treated patients develop post-operative complications such as endoleaks, endograft migration, or limb occlusion, requiring reinterventions2. The latter is a major complication as it causes claudication and acute limb ischemia, which often requires reintervention3. The Anaconda endograft (Terumo Aortic, Inchinnan, Scotland, UK) has been related to higher limb occlusion rates, despite its kink-preventing and multiple independent ring design4. This may be an unforeseen side effect of the Anaconda's flexibility crib/valley ring design which can be expected to disturb the blood flow dynamics (e.g., fluid stasis and recirculation zones); the so-called Concertina effect4. These unfavourable hemodynamic parameters may increase the platelet activation potential (PAP), which could eventually promote limb occlusion. Method: A thin-walled AAA flow phantom was fabricated based on a patient's anatomy, who was treated by an Anaconda endograft. This phantom was designed using the latest post-operative computed tomography angiography (CTA) scan of the patient just before an occlusion occurred in the left graft limb. We performed contrast-enhanced ultrasound particle image velocimetry (echoPIV) measurements to quantify the velocity fields. The echoPIV measurements were performed in the same phantom with and without Anaconda endograft to investigate the rings' effects over the local flow field. Finally, vector complexity (VC), vortices, and particle residence time were calculated. Results: The iliac velocity fields obtained from the pre-operative measurements had low VC values (below 0.2). On the contrary, the post-operative measurements showed locally high VC values, especially near the stent ring. Furthermore, we observed flow complexity zones (e.g., vortices and flow stasis), mainly in the left graft limb, which may justify the detected patient limb occlusion. Conclusion: In this feasibility study, we identify and define unfavorable hemodynamic parameters that might promote PAP, and therefore carry a great potential as prognostic value for limb occlusion. Future studies must be done to further generalize the hypothesis and evaluate the defined hemodynamic norms. 16:00 0 mins Multi-perspective photoacoustic imaging compounding based on a deep learning approach Amir Gholampour, Navchetan Awasthi, Jan-Willem Muller, Hans-Martin Schwab, Min Wu, Josien Pluim, Richard Lopata Abstract: Photoacoustic imaging has great potential in providing information about tissue morphology. Linear or curved arrays are commonly used to perform this imaging technique. However, the resolution, contrast, and field of view are limited. Multi-perspective photoacoustic imaging (MP-PAI) can overcome these limitations by increasing the area for receiving photoacoustic signals using multiple transducers. The conventional approach to fuse the images from individual transducers is to average, which is simple and straightforward, however, this is not always optimal. In this work, we propose to employ a deep neural network using a modified U-net architecture to improve the image fusion step (compounding). To generate the dataset, a simulation framework is implemented which consists of a Monte-Carlo step to simulate the optical fluence and the k-Wave toolbox to simulate acoustic wave propagation. In the simulated phantoms, the optical and acoustic properties are randomized. The positions and orientations of the transducers and the light sources are randomized as well. The dataset contains 1400 set of images that are separated into 1000 for training, 200 for validation, and 200 for testing. The compounded images, in general, resemble the initial pressures that were used for the simulations, revealing great improvement in terms of resolution and contrast. To compare the results of both conventional and deep learning-based compounding approaches against the initial pressure, SSIM and PSNR evaluations are used. The average SSIM improved from 0.08 in conventional compounding to 0.84 in the deep learning approach, and the average PSNR improved from 5 dB to 30.14 dB. Overall, the preliminary results are promising and demonstrate the feasibility of this approach for employment in the experimental framework. In future studies, the training dataset can be extended by increasing complexity of the phantoms to improve the robustness of this method. 16:00 0 mins Skateboard parameter optimization for the ollie with optimal control using direct collocation Jan Thomas Heinen Abstract: Skateboarding involves a human controlling a four wheeled vehicle that is steered by tilting the standing surface. The riding mechanics of skateboarding have been well reported [1][2]. The sport also includes aerial maneuvers such as jumping of stairs, flying off ramps and flipping and rotating the skateboard. The most basic aerial trick is called the ollie. The athlete jumps up while pushing down on the back end of the skateboard’s tail, causing a rotation about the back axle. The upward acceleration due to the rotation together with the tail-ground impact cause the skateboard to go airborne. Midair the athlete pulls the skateboard up through frictional contact and levels it out to land the trick. The most absolute performance measure of the trick is height [3]. To reach maximum height the dynamics such as impact, dynamic response, and torque production are dependent on shape, inertia and mass, which gives reason to assume an optimal shape exists. This leads to the research question: How do geometric parameters influence the maximum jumping height during the ollie. A 9-piece two- dimensional model is made with a theoretical inertia values. The inertia values are tested and scaled appropriately to verify the model. With a multi-phase direct collocation scheme the model is optimized for maximum height subject to the dynamics of the skater and human to find the optimal trajectory and parameters. The ollie height is improved by changing either one of these parameters relative to a standardized skateboard; longer wheelbase, smaller tail angle, shorter flat part, and lower truck height. The results could be beneficial for Olympic performers to score higher points and push the sport to a new level. The method does not only apply to skateboarding, but any maneuver involving a human used artefact which has performance related dimensions. Such as, golf, honk ball, or cycling could be optimized similarly; model the artefact, simplify the input, set a performance metric, and optimize for optimal control and dimensions. 16:00 0 mins Protocol of pilot study: a serious game for cervical dystonia Luis Felipe García Arias, Martje van Egmond, Marina de Koning-Tijssen, Elisabeth Wilhelm Abstract: Cervical Dystonia (CD) is a life-lasting movement disorder causing depression, anxiety, and loss of self-confidence [1]. Treatment of CD requires an intense rehabilitation program, which is expensive and difficult to conciliate with daily living activities [2]. Since feedback is absent during home-based therapy, the patient could perform unintended movements and affect the rehabilitation outcomes. Performing the wrong movements can lead to a lack of progress perception, and the patient can feel demotivated to follow the treatment. Positive reinforcement and motivational strategies can be mediated by a videogame in addition to feedback on exercise execution [3]. Moreover, improving home-based exercises could reduce costs to the health system, enhance adherence and improve therapy outcomes [4], [5]. This study aims to investigate the feasibility of home training using virtual reality and the optimal number and location of sensors for providing feedback based on the movements and muscular activity of the patient. A prototype of a serious game to rehabilitate CD, which introduces the exercises and guides the user towards the desired position, will be developed. The prototype will give visual feedback based on the movement of the user. Five adults diagnosed with CD will be recruited from the outpatients of the University Medical Center of Groningen (UMCG). Inclusion criteria are stable treatment based on botulin neurotoxin for at least a year and no neurological comorbidities. After signing informed consent, basic demographic information will be recorded. While playing the game, the participants will use Inertial Movement Units (IMU) located in the forehead, the first and seventh cervical vertebrae, both scapulae, and the eighth thoracic vertebra. In addition, sternocleidomastoid, spinalis, levator, splenius capitis, and trapezius descendens muscles will be monitored using surface electromyography (EMG). The game will consist of following a bird in a virtual forest; the path the bird flies will be set according to the direction of the patient’s dystonic movements. After the game, the Active Range of Movement will be recorded, and the Intrinsic Motivation Inventory and information about the user’s experience will be assessed with 7-point Likert scales. The data will be used to identify the locations of both IMU and EMG that contain the most information about movement and muscles activation to give feedback. The subjective feedback will be used to improve the gaming experience. 16:00 0 mins Effect of patient population and electrode location on the transferability of an automated EEG sleep staging model Jaap van der Aar Abstract: Sleep stage scoring, an essential component of sleep disorder diagnostics, classifies 30-second recordings as either Wake, N1, N2, N3, or rapid-eye-movement (REM) sleep. Conventionally, polysomnograhy (PSG) recordings are manually scored by clinicians, a labour-intensive process which can suffer from inter-rater variability. In the recent years, a plethora of automated neural networks for sleep scoring have been developed for single-electrode electroencephalography (EEG), including SleepNet, SeqSleepNet, and DeepSleepNet. Although showing high performance, training of these models is often limited to specific electrode locations and large heterogeneous datasets of young, healthy subjects. In contrast, the clinical practice deals with a range of sleep disorders, heterogeneous patient characteristics, and a limited amount of data. Also, recordings from the specific electrode location are not always available, either because of a different setup or bad signal quality. Therefore, we study to which extent these well-trained models can be utilized when data characteristics change. First, TinySleepNet (a computational lighter version of DeepSleepNet) is trained on a dataset of 94 healthy subject, using the F3-M2 EEG electrode derivation and 10-fold cross validation. Next, we repeat the experiment while changing either the electrode location (C3-M2; Cz-Fz; F3-F4), the population (age-matched mild-to-moderate obstructive sleep apnea; age-matched psychophysiological insomnia), or any combination of above-mentioned electrode locations and populations. For each changing characteristic, we study whether the pre-trained model can sufficiently predict sleep stages in the new data, and if not, whether partial retraining with a limited amount of new data is possible, or full retraining of the model is required. Results show to which extent a widely used sleep stage model, trained on a healthy and heterogenous population, can be used in the clinical setting. We show under which conditions the model can be used directly, when a small sample of additional data is required, and when full retraining is necessary. Furthermore, the implications of this study can pave the road for more accurate automated sleep staging in new wearable EEG devices, where electrode locations differ, signal quality can suffer, and data availability is limited. 16:00 0 mins Human learning and generalization to stand with unexpected sensorimotor delays Brandon Rasman, Jean-Sebastien Blouin, Patrick Forbes Abstract: Sensorimotor delays are a ubiquitous feature of movement control in all animals. Consequently, the brain must learn to control movement based on outdated sensory information and compensate for the influence a delay may have on volitional self-motion. Sensorimotor delays for human standing balance can be particularly long (up to 160 milliseconds) and are exacerbated through aging and neurological diseases. Failure to accommodate for these delays will result in instability and falls. Computational models of human balance have suggested that we cannot maintain upright balance with delays longer than 300-340 milliseconds. Using a custom-designed robotic balance simulator, we tested whether healthy volunteers could learn to balance with control delays longer than the predicted 300-340 millisecond limit. This unique device allowed us to artificially impose delays into the control of standing balance by increasing the time between the generation of motor signals and resulting whole-body motion. The experiments demonstrated that lengthening the sensorimotor delay initially destabilized upright standing, but through training, participants regained the ability to balance. We further tested whether humans can generalize this learning across different contexts (i.e., varying imposed delays, balancing in different directions, balancing with different muscle groups). After training in a single condition (i.e., single delay and balance direction), participants exhibited balance improvements in both trained and untrained contexts, demonstrating generalization of learning. Finally, a subset of participants was tested three months later and were still able to compensate for the increased delay, demonstrating learning retention. Our results reveal that while long delays are initially destabilizing for human standing, the brain can learn to overcome delays up to 560 milliseconds in the control of balance. Furthermore, the brain can generalize these learned control principles to balance upright across varying sensorimotor delays, movement directions and mechanically-independent motor effectors. Our findings may have important implications for people who develop balance problems due to older age or neurological diseases like diabetic neuropathy. It is possible that robot-assisted training therapies, like the one used in our study, could help people overcome their balance impairments. 16:00 0 mins Cervical cancer brachytherapy: Time-action and patient experience analyses Sharline van Vliet - Perez, 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 (LACC). However, the patient’s experience in terms of pain, anxiety, and duration of each BT procedure step is still scarcely reported. The aim of this prospective study is to perform systematic time-action and patient experience analyses during cervical cancer BT as a benchmark for new technological developments and to guide further improvements. Method LACC patients treated with high-dose-rate BT were included for a time-action analysis (56 patients, 135 fractions) and patient experience analysis (29 patients, 70 fractions). 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, 100= best imaginable health) at the beginning of the day to establish a baseline 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 (IQR) for all parameters is reported. Results The total procedure time (hours:minutes) from arrival at the BT department till discharge was 8:50 (8:00-9:25), i.e. preparation implantation, applicator implantation, recovery from implantation, imaging, treatment planning, irradiation, applicator removal, and recovery. Treatment planning was the longest treatment step with a time of 02:55 (IQR: 02:25-03:15). At the beginning of the day, patients had a health state index of 0.80 (IQR: 0.65-0.89) and EQ VAS score of 70 (IQR: 55-80). During treatment, the highest pain scores were reported while waiting for irradiation during treatment planning (median: 3, IQR: 0-6) and applicator removal (median: 3, IQR: 1-7), the highest anxiety scores were reported during applicator removal (median: 2, IQR: 0-7), and the highest perceptions of duration were reported during image acquisition (median: 4, IQR: 0-6). Conclusion This analysis highlights patient experience during different steps of the LACC BT workflow. The time-action and patient experience analyses can be used to optimise different steps of the BT treatment and as baseline for future studies. 16:00 0 mins Patient-reported bicycle use and mobility of vulvar cancer patients Nick van de Berg, Franciscus van Beurden, Heleen van Beekhuizen, Marianne Maliepaard, Helena van Doorn Abstract: Background: Bicycling is an integral part of Dutch life, which facilitates nearly a quarter of all our journeys. As an urban traffic alternative, it has economic, social, and health benefits, and can potentially reduce traffic congestion and air and noise pollution. The societal value of enabling people to bicycle is recognised globally, and becomes visible from the multitude of bicycle-sharing systems and bicycle highways introduced in major cities around the world. Medical and technical universities can contribute to this enablement. However, to be effective, we need to identify population subgroups and investigate prevalence of bicycling impediments. Based on signals from our own patients, we have decided to investigate the effects of vulvar cancer and its surgical treatment on bicycle use, saddle pain and discomfort, and on overall losses in mobility and quality of life (QoL). Methodology: Patients who underwent vulvectomy in our hospital between 2018-2021 were retrospectively asked to complete the EQ-5D-5L and EQ-VAS questionnaire to estimate quality of life and perceived health, and a non-validated, problem-specific GO-bicycling questionnaire to assess perceived problems during bicycling. The study was approved by the Erasmus MC MERC and all participants gave consent before filling in the questionnaires. Results: In total, 84 patients (63%) filled in the questionnaires. Respondents had a mean age of 68±12 years (mean ± standard deviation). We found that QoL was similar to Dutch baseline values (0.832±0.224 vs 0.858±0.168), perceived health was reduced (75.6±20.0 vs 83.2±11.8), and mobility problems were doubled (31.0% vs 16.0%). The average bicycling time was also reduced (96±169 vs 194±257 min/week). Overall, 40.3% of respondents experienced moderate or severe bicycling problems or could not bicycle, 34.9% felt impeded to bicycle because of their vulva, and 57.1% wished to be able to make more or longer bicycling journeys. Chafing and pain in the vulva or sit-bones were the most frequently indicated types of discomfort. Conclusion: Our study shows that pain and discomfort caused by vulvar cancer and its treatment strongly reduces bicycle use and mobility. This motivates us to investigate ways to reduce bicycling pain and discomfort and improve mobility and self-reliance of these women. 16:00 0 mins Augmented feedback serious game platform for lower-limb rehabilitation Yiling Zhang, Hans Timmerman, Ming Cao, Elisabeth Wilhelm Abstract: Total knee arthroplasty (TKA) has good clinical outcomes. However, a significant proportion of the patients experience chronic pain [1], which is due to the changes that occur in the central nervous system (CNS). When the pain persists, reorganization in the brain may have actually contributed to chronic pain [2]. Sensory-retraining, such as an exercise conditioning program, might help patients after TKA return to daily activities and reduce their chronic pain. Therefore, we developed a study protocol of designing and evaluating a serious game platform for patients who suffer from chronic pain after knee surgery. There are a series of rehabilitation training games in the developed serious game platform. In the rehabilitation process, a time of flight camera (Azure Kinect) and inertial measurement units (IMUs) are used to detect and record exercise movement trajectories of players. The game platform scores the player’s game performance by comparing the player’s movement trajectory with the standard movement trajectory. This platform not only guides patients to exercise movements correctly, but also provides augmented feedback (AF) to improve their proprioception for enhancing the training effect. For example, the movement sonification maps player’s real-time movement data onto psychoacoustic parameters (i.e., loudness, pitch, timbre, harmony and rhythm) in order to guide the player to adjust their movements as standard rehabilitation exercise as possible; The game avatar on the screen generates visual feedback which reflects the movement performance of the player. Thus, the influence of AF on the effect of rehabilitation training is also investigated. A randomized controlled trail (RCT) test is conducted to validate whether the proposed rehabilitation exergame platform with AF provides more advantages. The reduction in the error between target movement trajectories and the practical movement trajectories of the participant is compared between the experimental group (within AF) and control group (without AF). Besides, the electromyography of 14 lower limb muscles, which are relevant to knee movement, are recorded by Delsys Trigno EMG sensors during the test to analyze the differences between the two group of participants’ muscle innervation after rehabilitation training. 16:00 0 mins Novel rank-based features of atrial potentials for the classification between paroxysmal and persistent atrial fibrillation Hanie Moghaddasi, Richard Hendriks, Alle-Jan van der Veen, Natasja de Groot, Borbala Hunyadi Abstract: Understanding the causes behind atrial fibrillation (AF) can be aided by employing intra-operative epicardial measures to examine action potentials (AP). Indeed electropathological characteristics derived from intra-operative epicardial measurements, such as conduction block (CB) and continuous conduction delay and block (cCDCB), can be used to assess the severity of AF. However, these parameters do not show a significant difference between patients having different degrees of AF while they are in sinus rhythm. The electropathology of atrial tissue can be comprehended from complementary features besides the parameters based on conduction velocity and wavefront propagation. It is important to note that beyond the wavefront propagation, there is an action potential that initiates and maintains the propagating wavefront. A wavefront may propagate normally even while the underlying APs are abnormal. In light of the possibility that variation in the atrial potential waveforms might be related to AF onset and maintenance, we develop a method in this study to assess the severity of AF. Using a singular value decomposition (SVD), we demonstrated that the spatial variation of atrial potential morphologies during a single beat can be expressed by two rank-based features proposed in this work. During sinus rhythm, we used 293 beats from patients who had experienced paroxysmal or persistent AF. Feeding a random forest classifier, we achieved 78.42% classification accuracy, while classification based on the CB and cCDCB led to an accuracy of 58.34%. 16:00 0 mins A model of heart failure patients for the generation of an in-silico cohort Hamed Moradi, Frans van de Vosse, Wouter Huberts Abstract: Implementing a pulmonary artery pressure sensor (PAPS) and continuous pressure monitoring improves patient management and significantly reduces HF patients' hospitalization rates [1]. The utility of PAPS is compromised by thrombus formation, device detachment, and migration. Therefore, precise device design and efficacy evaluation are of paramount importance. The use of virtual patient cohorts based on realistic cardiovascular models can help the verification and validation of the sensor because model parameter variations can create a large number of virtual patients that cover a wide variety of patients' phenotypes [2], something that can hardly be achieved in real clinical trials. The aim of this study was the development and validation of a model of an HF patient that realistically mimics the hemodynamics in the pulmonary both with and without a sensor implanted and can serve as the basis for a virtual cohort generator. To create the model, a 3D geometry of the pulmonary artery was coupled to a 0D lumped parameter model, which enables us to have a realistic simulation of PAPS. The 0D model estimates the human cardiovascular system's physiological flow and pressure waveforms at the boundaries of the 3D domain, whereas the 3D model gives a detailed picture of the local pressure and velocity fields around the PAPS. The parameters of the left and right ventricles' lumped model are changed to simulate the sick heart and the resulting pulmonary hypertensive situation. Validation of the simulated hemodynamics and the PAP sensor with literature and experiments showed a good alignment with them. For the evaluation of the PAPS, the output of interest can be listed as the pressure at the sensor's location, wall shear stress, the net force on the sensor, and the flow distribution through the side branches of PA. The developed model enables us to investigate all physiological and non-physiological parameters in cardiovascular patients and devices. Moreover, it can now be used to generate realistic virtual patient cohorts. 16:00 0 mins An anthropomorphic thyroid nodule phantom for testing novel needle-based interventions Tim Boers, Sicco Braak, Michel Versluis, Srirang Manohar Abstract: Purpose (main research questions) Thyroid nodules are benign tumors that can cause compressive symptoms and cosmetic issues, this affects roughly 5% of the adult population. In an increasing number of hospitals, a minimally invasive treatment pathway is offered. First, a biopsy is taken and thereafter a radiofrequency (RF) ablation is performed. However, these needle-based interventions suffer from a lack of precision leading to a lower diagnostic yield, incomplete treatment and regrowth of the nodule over time. Improving these needle-based interventions may solve that issue. However, testing this on patients is not without risks. In order to facilitate that improvement process without burdening patients, an anthropomorphic thyroid nodule phantom was developed. Additionally, such a phantom may also be used for training purposes. Method A human neck MRI scan (N=1) was used to create the digital models of the thyroid, blood vessels, trachea and nerves. These models were then 3D printed. Thereafter, a gel was created using polyacrylamide, water and egg-white extract, of which the latter is used to visualize the ablated area on a T2-weighted MRI scan. The acoustic attenuation, thermal properties, density, electrical conductivity and temperature correlation with the T2-MRI signal of the gel were characterized. The phantoms are placed in a set-up with a heated saline flow (37 degrees Celsius). To test the entire set-up its feasibility and utility, three phantoms were treated using radiofrequency ablation. Results The gel its physical characteristics, when compared to those of human thyroid, showed good correspondence. The size of the ablated area matched that of the MRI scan. The temperature overlays showed a good match with the ablated areas on the MRI scan. The impact of moving fluid near the ablation area was visible. Conclusion (general significance) An anthropomorphic phantom capable of mimicking the human neck, including ‘blood’ flow was developed. This phantom is suitable for the development and assessment of needle-based interventions for thyroid nodules without risk or burden to patients. Moreover, as the ablation area can be evaluated objectively, this phantom can also be used for training purposes, allowing clinicians to speed up the learning curve for thermal ablations. 16:00 0 mins An average model with a realistic geometry of abdominal aortic aneurysms Jeffrey Nagel, Hadi Mirgolbabaee, Michel Versluis, Michel Reijnen, Erik Groot Jebbink Abstract: An abdominal aortic aneurysm (AAA) is a permanent dilatation of the abdominal aorta. Depending on the size of the aneurysm and its growth rate, treatment varies from watchful waiting to surgery. With the development of new and advanced imaging and treatment techniques for AAAs, physical and digital vascular phantoms have become an important tool for validation and training. However, current phantoms are typically based on very basic models, often without a relevant physiological geometry of the branching vessels, or they are patient-specific, limiting the generalizability of results. Having access to average models with a more realistic geometry could prove beneficial in making more accurate phantoms and would allow investigation and comparison of anatomical differences between patient groups, linked to clinical outcome after endovascular aneurysm repair (EVAR). Here, we develop an average model with a realistic geometry of the AAA. The abdominal aorta was segmented from CTA scans of 64 patients that underwent elective EVAR. For each patient, two segmentations were made using Materialise Mimics software; one of the flow lumen and one of the aortic wall, including the mural thrombus. The centre lumen line (CLL) was extracted from the segmentations and all CLLs were registered to the same coordinate system. The CLL was divided into sections based on the branches of the abdominal aorta – aorta, renal arteries and common, internal and external iliac arteries – and each section was sampled at a constant sampling rate between segmentations. At each sampling point the cross-sectional area of the lumen was determined, from which the lumen diameter was calculated. The CLL coordinates and the corresponding lumen diameters were averaged at each point, and computer-aided design software (SolidWorks) was used to convert them into an average 3D model of the abdominal aorta. A phantom was 3D printed for future in vitro experiments, using flexible 80A resin from Formlabs. In future studies we will also make models of specific subsets of patient groups, based on treatment outcomes, using principal component analysis to investigate whether geometrical features can be used to predict potential complications after EVAR, such as limb occlusion and type I endoleak. 16:00 0 mins BECA: Activating the Chain of Survival for Unwitnessed Out-of-hospital Cardiac Arrest Roelof G. Hup, Emma C. Linssen, Lukas R.C. Dekker, Reinder Haakma, Hanno L. Tan, Rik Vullings Abstract: Out-of-hospital cardiac arrest (OHCA) is a major health problem and occurs in individuals of all ages, sexes, ethnicities, and socioeconomic positions. Within the European Union, the incidence is estimated to be around 343.000 cases per year [1], with an overall survival rate of 10% [2]. It is essential that the chain of survival, i.e., alarming and early basic and advanced life support, is started as soon as possible. Unfortunately, 40% of the cases has no witness to start this chain of survival [3], which decimates the chance of survival. In our project, BEating Cardiac Arrest (BECA), we aim to develop a prototype device that can detect OHCA and start the chain of survival, even when no witnesses are present. By using a smartwatch that measures blood volume (using photoplethysmography) and accelerometric data at the wrist, we aim to detect circulatory arrest. After OHCA detection, GPS location can be sent over a mobile connection to a Dutch citizen-responder system, which will act as a link to citizen-responders and emergency medical services. Several clinical studies will be conducted to acquire data that is used for development and validation of the OHCA detection method. In the first study, 60 healthy volunteers will be asked to simulate OHCA several times by stopping movement and inflating a blood pressure cuff on the side where the smartwatch is carried. Secondly, a group of 30 patients with an implantable cardioverter defibrillator (ICD) is followed over a longer period to capture clinically occurring OHCA episodes. Lastly, 50 patients will be studied during either a ventricular tachycardia (VT) ablation or a subcutaneous ICD implantation, in which ventricular tachycardia/fibrillation is induced as part of clinical practice. With these groups, we aim to approximate the group of OHCA victims as good as possible. Other works will include a target group study, as well as studies into the ethical and psychological aspects of the technology. By investigating the technical, clinical, societal, economic, psychological, and ethical aspects, this project will increase the chance of adoption by the medical community. For this reason, a consortium is formed consisting of academic partners (from the areas of cardiology, engineering, psychology, behavioral sciences, ethics), a large MedTech company, and a technology provider for citizen-responder systems. REFERENCES [1] Empana, J.-P., Lerner, I., Valentin, E., Folke, F., Böttiger, B., Gislason, G., Jonsson, M., Ringh, M., Beganton, F., Bougouin, W., Marijon, E., Blom, M., Tan, H., & Jouven, X. (2022). Incidence of Sudden Cardiac Death in the European Union. Journal of the American College of Cardiology, 79(18), 1818–1827. [2] Myerburg, R. J. (2014). Initiatives for improving out-of-hospital cardiac arrest outcomes. Circulation, 130(21), 1840–1843. [3] Straus, S. M. J. M., Bleumink, G. S., Dieleman, J. P., Lei Van Der, J., Stricker, B. H. C., & Sturkenboom, M. C. J. M. (2004). The incidence of sudden cardiac death in the general population. Journal of Clinical Epidemiology, 57(1), 98–102. 16:00 0 mins Predicting Uncertainty of Metabolite Quantification in Magnetic Resonance Spectroscopy with Applications for Adaptive Ensembling Julian Merkofer, Sina Amirrajab, Johan van den Brink, Mitko Veta, Jacobus Jansen, Marcel Breeuwer, Ruud van Sloun Abstract: Current deep learning methods for metabolite quantification in MR spectroscopy do not offer reliable measures for uncertainty. Having such a measure would not only flag potential (self-identified) fitting errors, but also enable uncertainty-based adaptive ensembling of classic model-based fitting and deep learning predictions. In this abstract, we propose a training strategy based on a log-likelihood cost that allows joint optimization of metabolite concentrations and uncertainty estimation for each individual metabolite. On synthetic data, we show that the predicted uncertainty correlates well with the actual estimation error and that uncertainty-based adaptive ensembling outperforms the individual estimators as well as standard ensembling. 16:00 0 mins 2D human pose estimation in the catheterization laboratory Rick Butler, Teddy Vijfvinkel, Sjors van Riel, Chavdar Bachvarov, Jan Constandse, Maarten van der Elst, John van den Dobbelsteen, Benno Hendriks Abstract: The field of workflow analysis aims to analyze limiting factors during medical procedures, and improve cost-efficiency and patient wellbeing [1, 2, 3]. We aim to automate Workflow Analysis in the Catheterization Laboratory (Cath Lab), where we recorded 300 Cardiac AngioGram procedures from 5 different viewpoints for this purpose. One feature that is very descriptive of personnel activity---and therefore workflow---are the positions and poses of people in the room over time [4, 5, 6, 7]. Therefore, we start by testing several pose detection algorithms [8, 9, 10] in a qualitative manner. We pick clinically interesting and technically challenging situations from our dataset and evaluate how each algorithm performs. We conclude with the up- and downsides of each algorithm, and a recommendation of which one to use in a clinical environment. 16:00 0 mins Neural MAP beamforming for ultrasound imaging Ben Luijten, Boudewine Ossenkoppele, Nico de Jong, Martin Verweij, Massimo Mischi, Ruud van Sloun Abstract: Ultrasound imaging is an attractive imaging modality due to its low-cost and real-time feedback, but often lacks in image quality as compared to MRI and CT imaging. Conventional ultrasound image reconstruction, such as Delay-and-Sum (DAS) beamforming, aims to find an optimal reconstruction x from measurements y, and is derived from maximum-likelihood (ML) estimation of the probability distribution p(y│x). As such, no prior information is exploited in the image formation process, which limits potential image quality. Maximum-a-posteriori (MAP) beamforming techniques aim to overcome this issue by including a prior distribution p(x). However, such methods often rely on rough approximations of the underlying signal statistics [1], or are prohibitively slow and complex for real-time imaging [2]. Deep learning based reconstruction methods have demonstrated impressive results over the past years, but often lack interpretability and require vast amounts of training data. Recently we have proposed a novel method, neural MAP, which efficiently combines deep learning in the MAP beamforming framework. In this framework we aim to overcome the challenges in MAP beamforming by learning the measured signal statistics, and the prior signal distribution, through neural networks. We show that this model-based deep learning approach can achieve high-quality imaging, improving over the state-of-the-art, without compromising the real-time abilities of ultrasound imaging. We acquired a train and test set containing distinct in-vivo images from different anatomies, through a Verasonics research platform in combination with a 128-channel 6.25-MHz linear array transducer. An 11 plane-wave (PW) imaging scheme was adopted, with transmitting angles equidistantly distributed across ± 18º. From these acquisitions we generated ground-truth images using a minimum-variance beamformer [3]. In this work we aimed to reconstruct these high-quality ground truth images from only a single 0º PW acquisition. We compared neural MAP against DAS (reference) and ABLE (state-of-the-art) [4], and evaluated the robustness of each method to input noise. To that end, we varied the input SNR from 0dB to 20dB compared to the original signal. On average we measured an increase of 9.9dB and 1.1dB in PSNR using neural MAP, compared to DAS and ABLE, respectively. 16:00 0 mins Reconstructing perceived faces from multi-subject fMRI activations with hyper-aligning and -decoding Thirza Dado, Yağmur Güçlütürk, Marcel van Gerven, Umut Güçlü Abstract: The field of neural decoding seeks to find which information about a perceived sensory stimulus is present in and can be retrieved from recorded brain activity. The adoption of generative adversarial networks (GANs) in neural decoding has been fruitful in that they effectively model the “synthesis” operation from feature space to real-world data samples. As such, it is possible to reconstruct GAN-synthesized face stimuli that were presented to participants based on their recorded neural activations. The close resemblance between the stimuli and their reconstructions from brain data indicates that GAN latent- and neural representations represent the images similarly. Here, we present how to more closely approximate the presented stimuli as perceived by two participants in the MRI scanner by hyper-aligning and -decoding their fMRI recordings. This approach enabled the training of a single general decoder model that captures the shared neural information of the two participants to more accurately predict the input stimulus. 16:00 0 mins Personalized FMRI encoding models using ANNs Dora Gözükara, Djamari Oetringer, Umut Güçlü, Linda Geerligs Abstract: Studying the feature maps of artificial neural network (ANN) models and their relation to the brain have led to many new insights in computational neuroscience. ANN features are being used in a variety of ways, from predicting neural activity to understanding neural computation. Nevertheless, making these features comparable with brain data is far from a straightforward task. Conventionally, features from an entire ANN layer are grouped together, regardless of their spatial selectivity or size, and are compared with spatially selective brain areas, such as the early visual cortex. Consequently, a large amount of data is needed to train successful brain encoding models with large parameter spaces, and these models in turn need to learn which spatial location is relevant for the voxel to which they are being fit. In this work, we are building personalized encoding models that predict voxel timeseries data from a movie-viewing fMRI dataset using ANN features. We use eye-tracking and retinotopic data to build our personalized models that are not only specific to each participant, but also to each participant’s voxels. We do this by combining eye-tracking and voxel population receptive field data to sample only the relevant parts of the ANN feature map at only the relevant timepoints in the movie. We show that our personalized models make it possible to successfully train encoding models using limited data by reducing the model parameter space. We also present a new way to structure ANN features so that their comparison with fMRI data becomes more straightforward. 16:00 0 mins New method in safe tissue handling; an innovatively balanced compliant laparoscopic grasper Jan-Willem Klok Abstract: Introduction In laparoscopy, surgeons occasionally find it difficult to perceive forces exerted to the tissue, as well as to assess tissue characteristics through haptic feedback. This is due to poor transmission of surgeons’ instrument control forces, caused by friction and play in the grasper joints. These factors negatively impact instrument haptic feedback and can cause excessive tissue damage. Therefore, a laparoscopic grasper overcoming these problems has been designed. Methods This laparoscopic grasper features a novel compliant grasper tip type. In contrast to conventional laparoscopic graspers, this compliant grasper tip does not have joints that can cause play or friction. Instead, it utilizes elastic elements facilitating the opening and closing movement of the jaws. The instrument’s sensitivity was validated by measuring the level at which participants could discriminate force differences applied at the grasper tip with an increasing force level, while the subjects were holding the new instrument and a conventional grasper at the handle grip. The force was increased and subjects reported at which force level they were able to feel an increase. Also, mechanical efficiency of both graspers was measured. Results There was a significant difference in sensitivity between the instruments (p<<0.05) in favor for the compliant grasper (mean 1.4N SD0.44N) compared to the conventional grasper (mean 2.7N SD1.2N). The mechanical efficiency of the new instrument and conventional grasper is 43% and 25%, respectively. Discussion and conclusion The results show that the compliant laparoscopic grasper has a higher force sensitivity and a higher mechanical efficiency than a conventional grasper, enabling to perceive lower tip forces. This has significant benefits, potentially lowering the threshold of tissue forces that a surgeon can perceive, improving the perception of grasper tip-tissue interaction. The instrument can help surgeons to discern healthy and cancerous tissue by perceiving stiffness. Based on these findings, a new prototype was built. It features an improved compliant tip with reduced actuation forces. Furthermore, instrument weight has been reduced. Currently, a new validation study on the haptic feedback is being developed, aiming to investigate the distinction between ‘self-generated’ haptic feedback from palpation, and ‘tissue-generated’ feedback from pulsating blood vessels. 16:00 0 mins Care for patients versus care for machines The effect of technology on the workload of intraoperative nurses, an observational study. Anneke Schouten, Steven Flipse, Diederich Cornelisse, Frank Willem Jansen, Anne van der Eijk, John van den Dobbelsteen Abstract: Background Over the years, technology has become an irreplaceable element in the operating room (OR). This technology predominantly focuses on improving patient safety, increasing time-efficiency and sometimes the well-being of the surgeons. The impact of technology on the nurses of the OR is, however, rarely investigated. At the same time, the shortage of nurses in hospitals is a problem felt all over the world. The influx of new nurses has decreased over the years, and amongst the current nursing staff there is a high turnover rate. Literature states that the excessive workload is one of the most pressing factors to counteract the aforementioned trends. It has been suggested that technology could play a role in reducing the workload. Aim and research question The aim of this article is to investigate how technology affects the workload of OR nurses, and how technology affects the working experience of OR nurses. We hypothesize there is a shift in the nature of the work of the OR nurses from care for the patient to care for the technology on the OR. Method In this research we make a distinction between the objective task load and the subjective task load. We also distinguish between medical tasks and technological tasks. Over the timespan of a year, the objective workload of intraoperative nurses will be mapped by filming their everyday work during surgical procedures. The research setup exists of two ORs and six cameras and we make use of Noldus Observer XT for annotation of the videos. The surgical procedures are divided in three levels of technology: the lowest level being open surgery (OS), level 2 is laparoscopic surgery (LS) and the highest-level being robot assisted laparoscopic surgery (RALS). The ratio of medical and technological tasks, and the height of the task load will be compared between the three types of procedures. The subjective task load will be mapped by the means of a questionnaire, based on the NASA-TLX workload measure. 16:00 0 mins Cloud-based ultrasound platform for advanced real-time image information Beatrice Federici, Ben Luijten, Andre Immink, Ruud van Sloun, Massimo Mischi Abstract: Background & Objective. Deep learning (DL) based signal processing methods are taking an ever more prominent role in ultrasound (US) imaging, demonstrating unprecedented opportunities for image formation and processing. Integrating these models in clinical devices poses additional challenges and increased cost due to their high computational footprint. Moreover, due to the interactive nature of US imaging, such a system should reliably operate at real-time frame rates and with minimal latency. This work proposes a cloud-based US platform that allows live streaming of raw, high-bandwidth, US channel data to the cloud and remote control of run-time parameters. The system enables inference of compute-intensive models for latency-sensitive applications. We assess the performance of the proposed framework by demonstrating real-time cloud-based US beamforming with a trained DL method. Methods. US channel data is acquired using a widely-used programmable US platform (Vantage 256, Verasonics). A linear probe (L11-4v) with 128 channels is adopted for plane wave imaging at 100 fps. The proposed framework includes the communication of US in-phase and quadrature channel data from the Verasonics system to Python on an on-premises server (replaceable by a public cloud). An optimized neural network for adaptive beamforming [1] processes the received channel data remotely and returns the reconstructed frames. After reconstruction, the frames can be sent back to the Verasonics system, rendered through a display window on a remote desktop, or visualized by a third-party connected device though a web application, (e.g., smartphone). A user interface enables control of transmit parameters. In a next step, parameters tuning can be made adaptive based on cloud-based learned algorithms, e.g., quality-based. Results & Discussion. Considering a single plane wave (243 KB) and assuming instantaneous processing, the framework introduces a 20 ms latency and an inter-frame period of about 15 ms. When integrating with the advanced beamformer, we measured a total latency between acquisition and rendering of less than 80 ms and a display frame rate of 20 fps. This work demonstrates the possibility to live stream high bandwidth US channel data over the network and opens up new opportunities for real-time cloud-based image formation and processing. 16:00 0 mins Pelvic floor muscle support in pelvic organ prolapse patients Frieda van den Noort, Mirjam de Vries, Luyun Chen, John de Lancey, Anique Grob Abstract: Introduction: Pelvic organ prolapse (POP) is the exposure of one or more pelvic organs outside the vaginal opening. POP is likely due to weakened pelvic floor muscles (PFM) and ligament support caused by vaginal delivery damage, aging, genetics or hormonal changes. We investigate the hypothesis that the PFM support from POP patients differs from controls. Furthermore, we investigate if the PFM support changes during the day. Some patients report symptom increase during the day, which we hypothesize might be due to changing PFM support. Methods: Upright 0.25T magnetic resonance imaging (MRI) data of 43 healthy volunteers and 15 POP patients was obtained in the morning, the early and late afternoon. PFM support was investigated by selecting 12 corresponding sagittal points on each MRI [1]. These points that reflect pelvic floor “sagging” were aligned and normalized and principal component (PC) analysis was applied to investigate shape differences, which allows us to analyse the PFM support by shape differences. The values of the first two PCs were used for statistical comparison between groups, using two-way ANOVA. Results: The first two PCs captured 61% and 33% of the shape variation. Both PCs showed significant difference between POP patients and controls (PC1 p<0.001 and PC2 p=0.049). No statistically significant differences were found at different daytime points for both PCs. Visual examination of PC1 revealed that it mostly captures the cranial/caudal shape variation, the average POP population shape is more caudal compared to controls, suggesting lower support. PC2 captures mostly the length shape variation, visual difference was minimal between groups. Conclusion: Visual and statistical comparison of the difference in PFM support between POP patients and controls strengthens the hypothesis that weakened PFM support is one of the underlying POP causes. No significant difference in PC shape is found during the day in both POP patients and controls. Comment: The PFM support may not explain the reported symptom increase later in the day in women with POP. However, the POP group in this study is small and not all patients report symptom increase. Further research including only POP patients reporting symptom increase is beneficial [1] P. Schmidt, L. Chen, J. O. DeLancey, and C. W. Swenson, “Preoperative level II/III MRI measures predicting long-term prolapse recurrence after native tissue repair,” Int Urogynecol J, vol. 33, no. 1, pp. 133–141, Jan. 2022, doi: 10.1007/s00192-021-04854-3. 16:00 0 mins Estimating power loss during wheelchair propulsion based on two inertial sensors and a mechanics-based machine learning model Marit van Dijk, Louise Heringa, Marco Hoozemans, Monique Berger, DirkJan Veeger Abstract: One important performance determinant in wheelchair sports is the mechanical power exchanged between the wheelchair athlete and the environment. Recently, we developed a feasible method to determine power in the field based on rolling resistance estimates. Although this method shows to be accurate, rolling resistance estimates are wrong for the first two or three pushes. Probably this is caused by upper body movements. The aim of current study was to develop a model that accurately estimates the instantaneous rolling resistance based on two inertial sensors during wheelchair propulsion on a treadmill. Twenty-five healthy participants performed five different 120s-trials in an instrumented sports wheelchair on a treadmill. The five trials differed with respect to mass (+0, +5 or +15 kg) or tire pressure (1.75, 3.5, 5.25 bar). Before each trial, body mass inertial parameters were obtained and wheel friction coefficients were determined based on drag tests. During the trials, inertial sensors were attached to the participants’ trunk (thorax) and wheelchair (wheel), upper body kinematics were monitored using an optical motion capture system and a load pin integrated in the front wheels of the wheelchair measured the instantaneous load on the wheels. Based on the instantaneous front wheel load and center of mass kinematics, rolling resistance was determined. Accordingly, a deep learning model was trained to estimate the proportion of front wheel load from inertial sensor data. The training set included data of 20 participants and three trials. The rolling resistance of the other participants and other trials was used to evaluate the model by comparing model estimates with actual instantaneous rolling resistance. Based on the four most predictive features from the inertial sensor data, a machine learning model was trained and evaluated. The root-mean-squared-error (RMSE) turned out to be less than 5% of the body mass for the participants that were excluded from the training set. For the trials that were excluded, the RMSE was slightly higher. Overall, the rolling resistance estimate was improved significantly. To conclude, by combining a machine learning model with existing mechanical models, the changes in mechanical power can be determined accurately based on two inertial sensors. 16:00 0 mins Improved rider-bicycle balance control with balance assist ebike Leila Alizadehsaravi, Jason Moore Abstract: A bicycle is more difficult to control at low speeds due to the vehicle's unstable low-speed dynamics. Factors such as aging, disturbances, and multi-tasking can excite instability. Together with the Royal Dutch Gazelle E-bike and Bosch eBikes system, we developed the second prototype 'balance assist system', at the Delft University of Technology. We implemented the full-state feedback control algorithm on the balance assist bicycle. We evaluated the effectiveness of the Balance assist on 18 old and 14 young cyclists in two single- and multi-task scenarios at a forward velocity between 1 to 5 m/s. In the single-task scenario, participants rode the bicycle on a 30 m straight-line track. In multi-task, in the middle of the track, they additionally performed lifting/placing one hand off/on the handlebar corresponding to the identified direction of the object at the start point following a shoulder check task. In total, participants performed 16 trials (Scenarios (2) x Balance assist on/off (2) x Disturbances on/off (2) x repetitions (2)). In half of the trials per scenario, we implemented three disturbances with random intervals through the steering motor to mimic a sudden push in single-task (0.5s) or a wind gust in multi-task (1s) disturbances. We calculated the bicycle's mean absolute roll and steering rate to investigate the effect of the balance assist system on these variables in all conditions. Balance assist improved the lateral stability, indicated by reduced mean absolute roll rate, in both scenarios and groups in the presence and absence of disturbances. Especially balance assist system imposed similar lateral stability in older to young riders in single-task cycling. Moreover, balance assist decreased the steering effort, indicated by decreased mean absolute steering rate, in single-task cycling in both age groups and the presence and absence of disturbances. When the balance assist was activated, the steering rate did not differ in multi-tasking. Overall, balance assist showed the potential to improve cyclists' stability in challenging conditions, such as multi-task cycling or being subjected to sudden disturbances. Balance assist also showed promising results in reducing the continuous steering effort, which leads to safer cycling.

end %-->