16:00   Poster Session 1 (Even numbers) Feasibility of 3D ultrasound strain imaging for uterine contraction visualization Anyi Cheng, Yizhou Huang, Lin Xu, Massimo Mischi Abstract: Research question Analysis of the uterine peristaltic activity in the form of periodic contractions has shown promise for the diagnosis of the uterine dysfunctions [1] and the prediction of fertilization outcomes [2]. However, current quantitative analyses primarily rely on two-dimensional transvaginal ultrasound (2D TVUS), which is hampered by a limited view of the uterus and unavoidable out-of-plane motion artefacts. Here we propose the use of three-dimensional (3D) TVUS to obtain novel, comprehensive assessment and visualization of the uterine contraction patterns and propagation-velocity field. Methods 3D TVUS data were acquired for 40 s from a healthy volunteer and an adenomyosis patient using a V5-9 probe at 5.6-MHz central frequency and 1-Hz volume rate. A geometrical model of the endometrium (EM) was determined manually based on the first volume and tracking markers (TMs) were regularly distributed over the EM surface with 10-degree intervals around the EM central line and 1 cm longitudinal spacing. Radial strain signals were derived over the EM surface by optical-flow speckle tracking. 3D spatiotemporal analysis of the radial strain was then performed over a time window corresponding to the mean contraction duration (Ct). Strain propagation velocity vectors were estimated over the EM surface by 2D cross-correlation of the radial-strain time-evolutions at neighboring TMs within the defined time window. Results and Conclusions The mean Ct was 24 s and 32 s for the healthy volunteer and adenomyosis patient, respectively. We derived the contraction propagation maps over a half-Ct time window for both subjects and showed the velocity vector fields estimated in the same periods. As expected, more irregular propagation can be observed in the adenomyosis patient. Indeed, 3D TVUS radial strain analysis can address the limitations of 2D acquisition and possibly provide new insights into uterine contraction propagation across the entire uterus, aiding in better comprehension of uterine activity and diagnostics. References [1] M. P. Andres, G. M. Borrelli, J. Ribeiro, E. C. Baracat, M. S. Abrão, and R. M. Kho, “Transvaginal ultrasound for the diagnosis of adenomyosis: systematic review and meta-analysis,” Journal of minimally invasive gynecology, vol. 25, no. 2, pp. 257–264, 2018. [2] F. Sammali, C. Blank, T. G. Bakkes, Y. Huang, C. Rabotti, B. C. Schoot, and M. Mischi, “Multi-modal uterine-activity measurements for prediction of embryo implantation by machine learning,” IEEE Access, vol. 9, pp. 47 096–47 111, 2021. Real-time intraoperative ultrasound registration for accurate surgical navigation in patients with pelvic malignancies M.A.J. Hiep, W.J. Heerink, H.C. Groen, L. Aguilera Saiz, B.A. Grotenhuis, G.L. Beets, A.G.J. Aalbers, K.F.D. Kuhlmann, T.J.M. Ruers Abstract: Surgical navigation aids surgeons in localizing and adequately resecting pelvic malignancies. Accuracy of the navigation system highly depends on the preceding registration procedure, which is generally performed using intraoperative fluoroscopy or CT. However, these ionizing methods are time-consuming and peroperative updates of the registration are cumbersome. In this present clinical study, several real-time intraoperative ultrasound (iUS) registration methods have been developed and evaluated for accuracy. During laparotomy in prospectively included patients, sterile electromagnetically tracked iUS acquisitions of the pelvic vessels and bones were collected. An initial registration and five other iUS registration methods were developed including real-time deep learning bone and artery segmentation of 2D ultrasound. For each registration method, the accuracy was computed as the target registration error (TRE) using pelvic lymph nodes (LNs) as targets. Thirty patients were included. The mean ± SD ultrasound acquisition time was 4.2 ± 1.4 minutes for the pelvic bone and 4.0 ± 1.1 minutes for the arteries. Deep learning bone and artery ultrasound segmentation resulted in an average (centerline)Dice of 0.85 and a mean surface distance below 2 mm. In 21 patients with visible LNs, initial registration resulted in a median (interquartile range [IQR]) TRE of 7.4 (5.9-10.9) mm. For the other five methods, 2D and 3D bone registration resulted in significantly lower TREs than 2D artery, 3D artery and bifurcation registration (two-sided Wilcoxon rank sum test p < 0.01). The real-time 2D bone registration method was most accurate with a median (IQR) TRE of 2.6 (1.3-5.7) mm. Real-time 2D iUS bone registration is a fast and accurate method for patient registration prior to surgical navigation and has advantages over current registration techniques. Because of the user dependency of iUS, intuitive software is crucial for optimal clinical implementation. Characterization of a REM-deprived phenotype in REM-dominant obstructive sleep apnea Luca Cerina, Pedro Fonseca, Gabriele Papini, Rik Vullings, Sebastiaan Overeem Abstract: Muscle atonia during REM is one of the factors affecting upper airway collapsibility in obstructive sleep apnea (OSA). Higher collapsibility can lead to more frequent respiratory events, sometimes negligibly (non-stage-specific OSA), sometimes significantly (REM-dominant OSA). The latter is associated with mild OSA (apnea-hypopnea index (AHI) 10-15), young age and female sex, but also with broadly heterogeneous symptoms not explained by typical OSA metrics, like the AHI. To improve its interpretability, the original definition (AHIREM / AHINREM > 2, AHI>5) was extended with various constraints (e.g., minimum REM time, maximum AHINREM ). However, this is a reductionist approach, and multiple REM-dominant sub-phenotypes may exist. The scope of this work is to explore the existence and the differences of two REM-dominant subgroups and non-stage-specific OSA. We deconstructed the AHI ratio formula as follows: AHIREM / AHINREM > 2 = (#eventsREM / #eventsNREM) * (timeNREM / timeREM) > 2 We then considered the general proportion of time in REM sleep compared to NREM (20-25% of total sleep) and how REM-dominance distributes along events and time ratios axes. Using an event ratio of 0.5 we separated the REM-dominance space in two subgroups, characterized by their relative REM time. Differences between subgroups and with non-stage-specific OSA were examined with Kruskal-Wallis test using Benjamini-Hochberg corrections in different clinical sleep datasets: SOMNIA (doi:10.1136/bmjopen-2019-030996), SHHS (doi:10.1093/sleep/20.12.1077) and MESA (doi:10.1093/aje/kwf113) with a sample size of 7132 unique PSG recordings. All datasets presented two subgroups. One with actual frequent REM events, the other presenting low REM (<15% of total sleep), which increased the relative weight of REM events in the formula. In a variety of polysomnographic and OSA metrics, and against actual REM-dominant OSA, the REM-deprived subgroup exhibited (among others): a higher AHI; more time awake after sleep onset (WASO) and subtracted disproportionally to REM; higher REM latency and lower sleep efficiency; more and deeper hypoxic events. The subgroup was significantly different both from non-stage-specific and actual REM-dominant OSA. We identified a subgroup of patients classified as REM-dominant OSA due to disproportionately low REM time, presenting unique phenotypical characteristics. This new subgroup may become a future target for personalized therapies. The effects of vibrotactile cueing of the foot sole on gait and its neurophysiological correlates during overground walking Ytjanda Sloot, Claudine Lamoth, Natasha Maurits, Menno Veldman Abstract: Parkinson’s Disease (PD) is a common neurodegenerative disorder, manifesting in multiple motor and non-motor symptoms. A frequently observed motor symptom is Freezing of Gait (FOG), which is an episodic involuntary inability to move. Cueing is a behavioral strategy that can be used to ameliorate FOG. Vibrotactile cueing, a cueing modality, has recently been applied in socks, but little is known about its effects on gait and the underlying neurophysiological mechanisms. External cueing is associated with gait improvements and altered cortical activity in healthy people and people with PD. Previously, cueing was found to induce age-dependent improvements in gait initiation and spatiotemporal gait parameters. Furthermore, electroencephalography (EEG)-derived metrics of plasticity, power and corticomuscular coherence (CMC), reflecting brain activity and brain-muscle connectivity, respectively, can be affected by cueing. Specifically, increased activation of sensorimotor brain regions and age-related increases in CMC are reported following external cueing in healthy people and people with PD. Thus, cortical sensorimotor activity and CMC may be important neurophysiological correlates that can improve our understanding of cue-induced gait improvements. Altered event-related brain dynamics during walking are previously shown in people with PD, as well as in healthy controls when they walk without arm swing. Vibrotactile cueing might ameliorate such altered brain activity. To examine this hypothesis, we investigate the effects of vibrotactile cueing on gait, cortical activity, and CMC during overground walking, both with and without arm swing, in healthy adults. Sixty healthy participants (18-30, 35-50, 60-75, 80+ years) will be included. Participants will complete 3-minute walking bouts: with and without arm swing, and with and without vibrotactile cueing applied to the foot soles through vibrating socks. 64-channel EEG data, electromyography data of bilateral gastrocnemius medialis and tibialis anterior and inertial measurement unit data of lower back, shanks and wrists will be collected to assess event-related spectral perturbation, CMC, and spatiotemporal gait parameters, respectively. Measurements will start in November 2024 and first results will be presented at BME2025. This study will enhance our understanding of how vibrotactile cueing affects gait and its underlying neurophysiological mechanisms in healthy adults. This knowledge potentially improves its application in various clinical conditions. Ankle Sensor-Based Detection of Freezing of Gait in Parkinson's Disease in semi-free living environments Juan Daniel Delgado Terán, Kjell Hillbrants, Dzeneta Mahmutović, ‪Tjitske Heida, Richard van Wezel Abstract: Introduction: Freezing of Gait (FOG) is a motor symptom experienced by people with Parkinson’s Disease (PD) where they feel like they are glued to the floor. Accurate and continuous detection of FOG is needed to implement effective cueing enabling it to prevent or shorten the episodes. Methods: A CNN model was developed to detect FOG episodes in data recorded from PD patients under free-living conditions. The data was divided into two datasets, the first set including all the movements and the second set including only walking and turning activities relevant to FOG detection. A data augmentation method was applied to the datasets (i.e. differential segmentation) to compensate for the episodic nature of FOG. The CNN model was evaluated using 5-fold cross-validation (5Fold-CV) and performance metrics such as accuracy, sensitivity, precision, F1-score, and AUROC. Results: Data from 24 PD participants was collected in a semi-controlled environment in ON/OFF medication. Data from three participants were excluded as they did not exhibit any FOG episodes. The model using only IMU data from the right ankle performed best (AUROC = 0.9596) when only walking and turning episodes were included, compared with All-Activities included (AUROC = 0.8888). Conclusion: The CNN model demonstrated excellent performance distinguishing between FOG and walking/turning activities. However, the performance diminished when all the activities were included due to variability in all the activities performed in the semi-free living environment, where sitting and standing were the activities most misclassified as FOG. Our results exhibit the challenge of differentiating between akinetic FOG and static positions. Implementing an activity threshold detection system to remove static windows could improve the model’s performance. Multilevel Data Fusion for atrial fibrillation detection in wearables Arlene John, Barry Cardiff, Deepu John Abstract: Background: Atrial fibrillation (AF), a common arrhythmia, can lead to serious health issues if undetected. Electrocardiogram (ECG) and photoplethsymogram (PPG) signals can be used to detect AF events, and fusion of these signals can improve detection performances. Traditional fusion methods rely on designer input and struggle to optimize fusion information abstraction levels under varying data quality. Integrating signal quality indicators (SQIs) into fusion models enhances reliability, yet adaptive SQI-based fusion for AF detection in wearable applications remains largely unexplored. Methods: A data-driven, multi-level fusion model that self-learns the optimal fusion stage for AF detection from ECG and PPG signals using 1-dimensional convolutional neural networks (1D-CNNs) is created. The fusion model incorporated SQIs per signal sample to prioritize cleaner signals during the fusion process. Using a subset of the MIMIC III database, 20-second non-overlapping windows of ECG and PPG data with associated SQIs were utilized. Four-streams of 1D-CNNs were employed: two streams each for ECG and PPG signals, and the individual SQI inputs for weighted feature fusion. A central fusion network was used to fuse the signals at varying stages in the feed-forward path based on the SQIs. The model was trained with two loss functions: combined network losses (average loss) and a central network loss. Simulated noisy signals were also used to assess the fusion architecture. Results: The model trained with average loss outperformed the central-loss-only model, achieving an accuracy of 99.33% accuracy and sensitivity of 99.74% under clean conditions. In simulated noisy conditions, the inclusion of SQI information improved accuracy by 3.80% compared to models without SQIs, demonstrating enhanced robustness. Compared to single-signal models, the multi-level fusion approach showed significant improvements, with accuracy gains of 3.51% over the ECG-only model and 14.55% over the PPG-only model. Conclusions: The fusion model provides a robust and flexible approach to AF detection in wearable devices by optimizing fusion levels through a data-driven approach and integrating SQIs for noise resilience. This self-learning, multi-level fusion architecture shows superior performance over existing fusion methods and holds promise for real-world applications in multimodal AF detection, potentially setting a new standard for AI-driven health monitoring systems. Understanding maternal-fetal cardiac coupling to improve perinatal and pregnancy outcomes Irene Lensen, Alessandra Galli, Elisabetta Peri, Massimo Mischi Abstract: Previous studies have shown the presence of cardiac coupling between fetus and mother [1]. Although this interaction is still largely unknown, pregnancy pathologies are associated with abnormal development of the fetal cardiovascular system (fCVS) and maladaptation of the maternal cardiovascular system (mCVS) to pregnancy [2]. Specifically, a difference in cardiac coupling was found between healthy and pathological pregnancies [3]. However, the impact of these mechanisms on pregnancy related complications is still uncertain. Better understanding of these mechanisms will contribute to improved monitoring of fetal development and early detection of pregnancy pathologies. To this end, investigating the cardiac coupling between the mCVS and fCVS during different gestational ages considering both healthy and pathological pregnancies could provide important insights. For this investigation, the HR series from mother (mHR) and fetus (fHR) are required and can be obtained from maternal ECG (mECG) and fetal ECG (fECG). These signals can be non-invasively derived from abdominal ECG signals, but the extraction of fECG is particularly challenging. This complexity arises from the small amplitude of fECG, which is masked by interference and noise sources that are also recorded by the abdominal electrodes and typically show higher amplitude. To obtain a reliable fECG estimation, a denoising approach based on the iterative combination of Periodic Component Analysis and Singular Value Decomposition has been developed [4]. Once mHR and fHR are extracted, the phase synchronization method can be applied to the time series to assess the phase coupling between mCVS and fCVS [5]. This method uses the relative mHR and fHR phases to quantify the coupling between the signals. When the relation between the relative phases is stable, the phase coupling index (i.e., a metric describing the amount of synchronization) is high, supporting the presence of coupling. To quantify the degree of synchronization between mHR and fHR, this index is calculated for the whole time series and periods of sufficient synchronization are determined using a threshold. Subsequently, the number and length of these periods are compared between different gestational ages and health statuses to identify significant differences, providing insights into the potential use of cardiac coupling for fetal monitoring. [1] Nichting, T. J. et al. (2023). Evidence and clinical relevance of maternal-fetal cardiac coupling: A scoping review. PloS one, 18(7). [2] Khlybova, S.V. et al. (2019). Heart rate variability in normal and complicated pregnancies. Hum Physiol 34. [3] Khandoker A. H. et al. (2019). Alterations in Maternal–Fetal Heart Rate Coupling Strength and Directions in Abnormal Fetuses. Front Physiol, 10. [4] Galli A. et al. (2024). Improved mECG Removal and fECG Extraction by Integrated Periodic Components Analysis and Singular Value Decomposition. 2024 IEEE International Symposium on Medical Measurements and Applications (MeMeA). [5] Wahbah M. et al. (2021). Estimating Gestational Age From Maternal-Fetal Heart Rate Coupling Parameters. IEEE Access, 9. Dynamic Steerable Patterning of Micro-Scale Particles and Living Cells using an Ultrasound Phased Array Rick van Bergen, Bart Groenen, Daniëlle Duffhues, Richard Lopata, Hans-Martin Schwab Abstract: Background, Motivation, and Objective Acoustic patterning is a non-contact method to pattern small scatterers, with innovative applications demonstrated in fields such as cell sorting, microfabrication, and tissue engineering. Patterning can be achieved by creating an ultrasound standing wave, generating a force toward the nodes. However, one downside of the conventional axial patterning is that a reflector-based or dual probe approach is required, limiting applications to very controlled setups and, in the context of tissue engineering, increasing the gap from in-vitro to in-vivo application. To expand on the current state of the art, we introduce a single-sided, lateral patterning approach exploiting the tunability of a clinical ultrasound transducer array that has the potential to in-situ pattern (engineered) tissues in a clinical setting. Statement of Contribution/Methods We show that the Verasonics P4-2v transducer array can induce acoustic patterning of silicon carbide particles in axial and lateral directions. To pattern laterally, we exploited the transmit beamforming capabilities of a transducer array to generate a Bessel beam. Adjusting the electronic transducer delays could dynamically alter the pattern's shape and orientation, which is impossible for single-element transducers commonly used for acoustic patterning. We also developed a numerical model to predict the experimental displacements by modelling our setup based on the K-wave toolbox. Results/Discussion We have shown that we can pattern particles using a clinical phased array. The simulated and experimental displacements have a slight phase mismatch (NCC = -0.12, Δk = 0.2 mm-1). The model accurately predicts the maximum displacement magnitude but overestimates the average displacement due to unaccounted friction. Furthermore, a phased array transducer allows for lateral patterning, with the displacement analysis confirming the simulation experiment agreement (NCC = 0.98, Δk = 0.02 mm-1). The steering capabilities of the ultrasound transducer array can be demonstrated by steering the pattern to angles of 6º and 21º. Finally, we worked on promising initial patterning results on living cells. To conclude, we used a clinical transducer array to pattern scatterers, augmenting patterning flexibility and enabling advances in the field of acoustic patterning and tissue engineering. The impact of voltage scaling on the energy efficiency in multichannel electrical stimulation applications Francesc Varkevisser, Tiago L. Costa, Wouter A. Serdijn Abstract: Electrical brain stimulation devices are successfully applied to treat several neural disorders, such as Parkinson’s disease, hearing loss, and visual impairment. Emerging applications, including visual and bidirectional somatosensory prostheses, drive the need for large-scale, fully implantable stimulator systems capable of interfacing with the brain through hundreds to thousands of stimulation channels. However, as the number of stimulation channels continues to scale, the available power becomes a significant bottleneck. Wireless powering is preferred, as it avoids the infection risks wired connections pose. However, the power that can be transferred to the implant is limited by several safety regulations. Consequently, optimizing the energy efficiency of stimulator circuits is essential to enable further channel scaling and ensure these devices can function effectively within the limits of available power. One way to improve energy efficiency is adaptive voltage scaling, where the power supply voltage for each stimulation channel is adjusted according to the tissue voltage requirements, defined by the electrode impedance and stimulation current. Several voltage scaling strategies have been proposed in the literature. This work reviews the effectiveness of different voltage scaling strategies across multichannel applications, including intracortical, retinal, and (intrafascicular) peripheral nerve stimulation, based on experimental data from existing literature. The analysis shows that for the conventional fixed voltage supply, efficiencies across the applications range between 35% and 54%, with power losses per channel ranging from 60 µW to 1 mW. Furthermore, the efficacy of the different scaling strategies varies across the applications. These findings suggest that tailored voltage scaling strategies could offer substantial efficiency improvements. By developing novel systems that support voltage scaling techniques, power efficiency can be enhanced, allowing for increasing the number of stimulation channels in next-generation, large-scale neural interfaces. Cross-Cohort Validation of the Advanced Alert Monitor: A Comparative Study in a Dutch Hospital Tom Bakkes, Ashley de Bie Dekker, Jonna van der Stam, Uzay Kaymak, Massimo Mischi, Arthur Bouwman, Simona Turco Abstract: Introduction: This study focuses on the generalizability of data-driven Early Warning Scores (EWS), which are used for risk assessment on the ward. One well-established score is the Advanced Alert Monitor (AAM) [1]. This data-driven EWS has been developed and validated in the United States. The AAM predicts the risks of unanticipated intensive care (IC) admissions and mortality within 12 hours. Prospective use of the score in Northern California led to a consortium-wide mortality reduction [2]. The success of this model has created interest in its generalizability to other hospitals. This study addresses this interest by validating the AAM overseas in a Dutch hospital. It examines how retraining the model and adapting the outcome definition affect the performance and feature importance. Method: Data was obtained from the Catharina Hospital's electronic health records. The AAM was reproduced based on the method of the original study [1] and locally retrained. A comparative analysis was performed between the original and local models. Additionally, two adverse event definitions were compared. The original definition was based on unanticipated IC admissions and mortality with a full-code care order. This means that mortality was not seen as deterioration if the patient did not consent to certain life-saving treatments. The second definition included mortality regardless of care order status. Result & Discussion: Results showed that local optimization of the model significantly improved the area under the precision-recall curve (AUPRC) from 10.1% to 16.2%. This suggests that local optimization of the model is recommended before implementation as it accounts for differences in population, clinical practices, protocols, and measurements. When training the model on the second outcome the performance increased even more to an AUPRC of 36.6% with an area under the receiver operating curve of 83.6%. A feature importance analysis showed that training on the second outcome focused the model more on physiological features than training the models on the original outcome. This indicates that the second outcome definition better captures patients showing physiological signs of deterioration, despite their care order status. These insights contribute to the ongoing development of data-driven EWS, intending to provide better clinical decision support. References: [1] P. Kipnis et al., Journal of Biomedical Informatics (2016) [2] G.J. Escobar et al., New England Journal of Medicine (2020) Ultrasound-based longitudinal study into hemodynamics and wall mechanics of abdominal aortic aneurysms Judith Fonken, Tessa Timmer, Arjet Nievergeld, Marc van Sambeek, Frans van de vosse, Richard Lopata Abstract: Biomechanical models are believed to improve rupture risk prediction in Abdominal Aortic Aneurysms (AAA), but their clinical translation is hampered by the use of costly and hazardous image modalities, such as CT and MRI. Three-dimensional time-resolved (3D+t) Ultrasound (US) is cost-effective, safe, and, could enable frequent use and large, longitudinal studies on AAA development. Recent improvements in US segmentation methods facilitate US-based biomechanical models to estimate and analyze biomechanical and hemodynamical parameters such as Peak Wall Stress (PWS), Peak Wall Rupture Index (PWRI), Time-Averaged Wall Shear Stress (TAWSS) and Oscillatory Shear Index (OSI). This study aims to investigate the influence of various biomechanical and hemodynamical parameters on AAA progression. Segmentations of the vessel wall and intraluminal thrombus (ILT) were obtained in a cohort of 30 patients with visible ILT and 9 patients with confirmed absence of ILT. Fluid-Structure Interaction (FSI) simulations were performed to calculate the hemodynamical and biomechanical parameters. Both qualitative and quantitative analyses were performed to investigate the relationship between size parameters (maximum diameter and vessel, ILT and lumen volumes, ILT thickness) and wall mechanics and hemodynamics (PWS, PWRI, TAWSS, OSI). Firstly, the mechanical parameters were investigated. This study showed that the positions of PWS, PWRI, and maximum ILT thickness may provide insights into growth rates. The distance between PWS and PWRI location significantly increased between slow- and fast-growing AAAs, with PWRI shifting towards maximum ILT thickness. In fast-growing AAAs, the ILT thickness at PWRI was found to be larger compared to slow-growing AAAs, even after correcting for the interaction effect by normalizing with maximum ILT thickness, considering that larger AAAs typically have larger ILTs. This study identified that vessel volume showed the strongest correlation with PWRI, while PWS exhibited a stronger correlation with lumen volume. Additionally, a new ILT size metric was introduced that demonstrates reduced dependency on AAA size compared to ILT fraction. This metric allows for a direct investigation into the effect of ILT on rupture risk, reducing the effect of the interaction between ILT and AAA vessel volume. In an ongoing study, the hemodynamical parameters are examined. Multi-modal imaging of stentgraft deformation and blood flow in aortic aneurysm phantoms Marleen Krommendijk, Hadi Mirgolbabaee, Elke Hestermann, Patryk Rygiel, Dieuwertje Alblas, Jelmer Wolterink, Erik Groot Jebbink, Michel Reijnen, Robert Geelkerken Abstract: Early postoperative morbidity and mortality in patients with an abdominal aortic aneurysm (AAA) have decreased significantly since the introduction of endovascular aneurysm repair (EVAR). However, complications after EVAR can develop over time, with major causes for reintervention being endoleaks related to AAA sac increase and occlusion of stentgraft limbs. Studies using echocardiogram-gated computed tomography (ECG-gated CT) scans of AAA patients treated with an Anaconda™ (Terumo Aortic, Inchinnan, UK) stentgraft have shown certain deformation patterns in the iliac stentgraft limbs over the cardiac cycle that could contribute to thrombus formation after EVAR. What still remains unknown is the relation between stentgraft deformation and blood flow. To gain insights into this relation, a study in patient-specific phantoms will be performed. The lumen and aneurysmal wall will be automatically segmented from dynamic computed tomography scans of thirteen AAA patients taken before stentgraft placement. Lumen development and thrombus wall deformation over the cardiac cycle will be analysed and visualised as deformation maps. The segmented aorta models will be 3D-printed to fabricate reproducible patient-specific phantoms that mimic the aortic deformation. A physiological pulsatile flow of blood-mimicking fluid will be generated through the phantom to complete the in-vitro setup. Blood flow patterns will be studied in this in-vitro setup before and after stentgraft placement. Global blood flow velocity and direction will be studied using four-dimensional phase contrast magnetic resonance imaging (4D-flow MRI). Echo particle image velocimetry (Echo-PIV) will be used for detailed localization of recirculation and stagnation zones and quantification of parameters related to wall shear stress and blood flow complexity. Stentgraft deformation patterns will be studied using ECG-gated CT by segmentation of the stentgraft rings and quantifying parameters such as inter-ring distance over the cardiac cycle. This in-vitro pipeline allows different AAA cases to be studied, along with comparison of different stentgraft types. The information gained through these experiments will contribute to the optimisation of stentgraft design to strive for a better long-term performance of stentgrafts after EVAR. Quantitative Analysis of Contrast Ultrasound Imaging for Characterization of Breast Tumors Florian Delberghe, Simona Turco, Zimei Lin, Pintong Huang, Massimo Mishi Abstract: Introduction: Breast cancer is the most common and second-deadliest cancer in women worldwide. Early detection and staging are decisive to choose the right treatment and improve survival. While mammography and breast B-mode ultrasound are standard diagnostic tools, they are limited in cases with dense breasts or isoechoic tumors. A common hallmark of cancers in different organs is chaotic neo-angiogenesis [1] that feeds the growth of the tumor. Contrast-enhanced ultrasound (CEUS) enables quantitative analysis of the vasculature, which can discriminate malignant tumors. This study aims to adapt the analysis of contrast dispersion in neo-angionenic vasculature [2], [3]– originally developed for organs like the prostate – to extract quantitative features from breast CEUS. These feature maps will be used to identify malignant lesions. Methods: Up until now, 50 patients underwent CEUS examinations with a bolus injection of SonoVue (Bracco) at the Zhejiang University School of Medicine (China). Patients with BI-RADS score 4A and higher [4] are then referred for small needle aspiration biopsy, which will serve as the ground-truth pathology. As breathing cannot be suppressed for the duration of the scan, the resulting motion is compensated using the iterative local search algorithm [5]. In cases where out of plane motion is too significant, or other issues affect the quality of the scans, they are rejected. Feature maps are then computed around the lesion delineated by the sonographer. Key features include parameters derived from time intensity curves (TICs) based on the fitted modified local density random walk (mLDRW) model [2] and similarity analysis between neighboring TICs [4]. Expected Results: Because the of the similar vasculatures of cancerous neoplasms between the breast and other tissues (e.g. the prostate), we expect similar trends in quantitative features. Specifically, cancerous lesions are likely to show increased spatial coherence between voxels, measured by correlation of the TICs and mutual information. For parameters derived from the mLDRW model, cancerous lesions are predicted to show higher and earlier enhancement combined with convection-dominated bolus dispersion. References: [1] B. P. Schneider and K. D. Miller, J. Clin. Oncol., 2005. [2] M. P. J. Kuenen, et al., Ultrasound Med. Biol., 2013. [3] C. Chen, et al., Eur. Radiol., 2024. [4] D. A. Spak, et al., Diagn. Interv. Imaging, 2017. [5] T. Tiyarattanachai, et al., Ultrasound Med. Biol., 2022. Towards a next generation bimodal stimulation electrode and nociceptive stimulator Frodo Muijzer, Remco Horstink Abstract: A recent report by Ipsos showed that chronic pain affects around 25% of adults in The Netherlands [1]. Pain related diseases are the leading cause of the disability and disease burden in the world [2]. Additionally, a survey in Europe showed that around 40% of chronic pain patients had inadequate management of their pain [3]. To move towards better solutions for chronic pain, a better understanding and characterization of pain and the underlying mechanisms is necessary. Nociceptive A𝛿- and A𝛽-fibers are responsible for detecting pain and touch stimuli respectively, and relay this information to the brain. In previous research, a bimodal stimulation electrode, using needle and disc stimulation, was used to selectively stimulate A𝛿- and A𝛽-fibers [4]. However, this electrode is now deprecated, and required components have been discontinued. A new bimodal electrode is necessary. Various options were investigated, and a new bimodal electrode was fabricated using a Flexible Printed Circuit (FPC). This allows for a mass-produceable electrode, small inter-electrode variability, and integration of the stimulation anode, where previously a separate Transcutaneous Electrical Nerve Stimulation (TENS) electrode was used. Initial tests show promising results, but a qualitative comparison with previous results is still to be performed. To support the new bimodal stimulation electrode in the current measurement setup, a new dual channel nociceptive stimulator is under development. The existing single-channel ambulant stimulator and eight channel stationary stimulator are combined into a new two-channel ambulant stimulator, upgraded with a modern Nordic nRF52 processor. This allows for new features such as Bluetooth Low Energy (BLE) with wireless synchronization, battery monitoring, and a compliance check. The latter allows to verify if the provided stimulus corresponds to the specified stimulus, and can warn if instability or lack of voltage resulted in an incorrect stimulus. Together with the move towards touch-proof connectors, these are major improvements regarding the safety of the new stimulator. Detecting and Monitoring Postpartum Uterine Contractions and Their Relationship to Postpartum Hemorrhage Using Electrohysterography Milad Mazaheri, Rik Vullings, Beatrijs van der Hout - van der Jagt Abstract: Postpartum hemorrhage (PPH) is one of the leading causes of maternal mortality, accounting for approximately 25% of all maternal deaths worldwide. Despite medical advancements, 14 million women experience PPH annually, leading to around 70,000 deaths. PPH is commonly caused by a remnant of the placenta after childbirth or ineffective contractions, the causality between the remnant of the placenta and contraction inefficiency can vary. When the uterine contractions are ineffective and inadequate following placental delivery, this could increase the risk of bleeding. Unfortunately, there is limited knowledge about the physiology of the uterus postpartum, due to the lack of accurate measurement tools. If postpartum activity of the uterus can be measured and interpreted correctly, it can ultimately improve the diagnosis and treatment of PPH. In this study, we propose the use of electrohysterography (EHG) for continuous monitoring of uterine contractions during the postpartum period. Abdominal EHG recordings collected during labor were extended postpartum by keeping the measurement electrodes on the abdomen after childbirth. In the initial phase of this research, we leveraged previous studies that successfully estimated Intrauterine Pressure (IUP) during antepartum contractions and propose to use this EHG-based IUP estimate to detect postpartum uterine contractions. The current phase of our research explores how these methods perform with postpartum EHG data and our preliminary findings indicate that it is possible to detect uterine activity postpartum with electrohysterography, offering the potential to enhance the assessment of uterine health after childbirth. Further investigation is required to improve the diagnosis and treatment of PPH, including integrating indicators of muscle fatigue alongside the proposed analysis of postpartum contractions. Accurately predicting the risk of PPH would enable early intervention and help prevent complications. This research paves the way for improved postpartum monitoring that allows earlier intervention and thereby contributes to improved maternal health outcomes and a reduction in PPH-related mortality worldwide. Maximum Information Line-scanning by Anatomical Manifold Tracking Wessel L. van Nierop, Oisín Nolan, Ben Luijten, Ruud J.G. van Sloun Abstract: Imaging rapid dynamics requires a high frame rate, in turn requiring a reduction of the number of transmit-receive events per frame. This typically manifests in a trade-off between image quality and frame rate. This work aims to reduce the number of measurements needed to obtain a high-quality ultrasound image by actively selecting the measurements that are expected to be most informative. To achieve this, we equip an imaging agent with a generative model of the ultrasound scene and observations, tracking beliefs about plausible anatomical explanations for the observations it performs. Based on these beliefs the agent pursues observations that have the highest expected information gain. We test our agent in a line-scanning mode, with the goal of reducing the number of transmit lines per frame. The agent utilizes a deep latent variable model, trained using variational inference on the CAMUS dataset, to operate on a low-dimensional latent manifold, enabling efficient inference and smooth frame-to-frame transition dynamics. On this manifold, a particle filter tracks the distribution over possible anatomical states given the observed scanlines. To select the next set of scanlines we perform maximum-entropy sampling, mapping latent particles into hypothetical future observations and using a multivariate Gaussian entropy model that includes line-to-line correlations to prevent naive sampling of redundant information. The figure displays two anatomies with proposed measurements highlighted in red. Reconstruction using subsampled measurements is shown in the second row, while the third row illustrates the variance of the distribution in image space. On the right, an illustration of the anatomical manifold tracking is given. The estimated posterior p(z_t│y_(1:t)) is updated over time using p(z_t│z_(t-1),y_t) where y_t are the proposed measurements. For selection of 16 out of 64 lines, active subsampling outperforms random and equispaced sampling by 10% and 4% respectively, with a mean squared error of 0.0103 ± 0.0036 for reconstruction performance. Exploring the predictability Prediction of Freezing of Gait and evaluation of the Parkinson's Vibrating Socks during daily-life activities in a house-like environment for people with Parkinson's disease Lorenzo Giuseppe Centamore, Marleen Tjepkema-Cloostermans, Richard van Wezel, Ciska Heida Abstract: Rationale: Freezing of Gait (FOG) is a troubling symptom in people with Parkinson’s disease (PD), disrupting mobility. Cueing, which involves external stimuli like auditory, visual, or tactile cues, shows promise in reducing freezing episodes. Developing a minimally invasive tactile system could overcome the usability limits of both visual and auditory cues. Cueing can be applied continuously or "on-demand," where on-demand cueing, triggered as needed, may prevent cue dependency and reduce fatigue, potentially offering longer-term effectiveness. To enable on-demand cueing, predicting FOG before it occurs is essential. Main Objectives: in this study, we aim 1) to study the feasibility of a FOG prediction algorithm under free-living conditions in a home environment, and 2) to evaluate the use of the Parkinson’s Vibrating Socks (a prototype of a minimally invasive tactile system positioned under the foot) under the same conditions. Data acquisition: Thirty participants with PD who experience daily FOG will be recruited. Data will be collected at the University of Twente (eHealth house) over one day. In the morning, participants will be monitored in their clinically defined OFF state, where FOG is more likely. In this way, symptom fluctuations will be recorded to develop a robust prediction algorithm adaptable to daily variations. Participants will wear motion, physiological (e.g., ECG), and eye-tracking sensors, while performing routine activities in a home-like lab environment, followed by a 280-meter indoor walking trail. In the afternoon, during their ON state, participants will walk the same trail under three cueing conditions: no cues, continuous cues, and on-demand cues (simulated) via the vibrating socks. Sock’s usability questionnaires will be filled in at the end. Data analysis: clustering techniques will be applied to outcome measures derived from the signals recorded using wearable motion sensors (e.g. gait, stress, environmental, and posture parameters), physiological sensors (e.g. skin conduction, heart rate, pupil diameter), and video evidence of FOG, to automatically label the time during which parameters starts to change before FOG (Pre-FOG). Subsequently, deep learning will be employed to build a predictive FOG algorithm. Qualitative and quantitative analyses of the vibrating socks’ usability will also be conducted, integrating sensor and survey data. Instrumenting aids for compliance and rehabilitation monitoring Kris Cuppens, Tessa Delien, Sarah Meyer, Veerle De Pourcq, Elvi Lemmens, Mario Broeckx, Tom Saey Abstract: Introduction: Thousands of people annually benefit from (custom-made) aids (e.g. orthoses, prostheses, wheelchairs, aids for activities of daily living, etc.). Miniaturized electronics and improved battery technology now allow for affordable, long-term monitoring sensor modules. Integrated into aids, these sensors can measure objective parameters, reducing fitting iterations, enabling real-time adjustments in rehabilitation, and detect issues preventively. This enhances the likelihood of correct and consistent usage, supporting treatment goals. This study investigates the conditions under which sensors can be integrated into aids, and how barriers can be lowered for therapists to be used in practice. Method: A survey with a patient association and a hospital assessed acceptance of sensors in orthoses and orthopedic shoes. Furthermore, several cases were investigated based on questions from (health)care organizations, orthopedic companies or technological companies. These included measuring therapy adherence, (peak) pressure, temperature, moisture and joint angles through sensors embedded in different (custom-made) aids. Commercially available sensor systems were tested when feasible. When such solutions were not available or too expensive, new demonstrators were developed. Sensor systems were first lab-tested, then evaluated by therapists and device providers in practical settings. Results and Conclusion: So far, 22 patients have completed the survey. Participants were asked about their acceptance of wearing orthoses and orthopedic shoes with integrated sensors. Results indicated 86% would wear sensor-equipped aids to improve therapy effectiveness, 73% would do so to enhance comfort, and 64% would accept it for monitoring adherence. When asked who should have access to the data collected, most respondents were willing to share data with their treating physician (77%), their therapist (73%), and the device provider (59%). Lab tests and try-outs with therapists showed that sensor systems effectively measured key parameters. However, challenges were identified, such as the cost of medically certified devices, choosing the appropriate sensor system amid numerous options, reimbursement for sensor integration, and adhering to MDR and privacy guidelines. Providing sensor information sheets and allowing therapists to experiment with various systems helps lower these barriers. Significance: This research advances the integration and acceptance of sensor-enabled aids to enhance therapy outcomes, improve patient comfort, and support data-driven healthcare practices. Review of existing clavicle loading analysis models towards reliable modelling methods of plate fixated clavicle loading during fracture healing. Annelies Verreth, Pieter Ansoms, Bryce Killen, Ilse Jonkers, Jos Vander Sloten Abstract: Introduction – Research question Clavicle fractures are the most frequent fracture within the active population, where surgical fixation is increasingly preferred to reach a faster return to typical function [1]. However, due to anatomical variability, standard fixation plates often mismatch anatomically, leading to soft-tissue irritation or plate removal after healing [2]. Personalization of fixation plates has been suggested as a solution to reduce these complications and associated revision surgeries [2]. To personalize plates to anatomy, the mechanical requirements to withstand expected loading during fracture healing must be known and subsequently estimated. Clavicle loading analysis through mechanical or in silico modelling is, however, limited in literature. Within this study, existing musculoskeletal models were compared to lead the way for development of a reliable model to analyze clavicular loading to inform personalized plate design. Research Method A literature review was conducted of the SimTk model database and the documentation of the AnyBody Managed Model Repository. The latter lists only one shoulder model, which is not freely accessible but is based on the Delft Shoulder and Elbow Model (DSEM) [4,5], which will be used for the comparison. In the SimTK database the musculoskeletal models containing the keywords ‘shoulder’, ‘upper extremity’ or ‘clavicle’ were selected for comparison if they were documented in a publication and validated for loading analysis. The selected models are the DSEM [4] and the Dynamic Upper Extremity model (DUEM) [3,6]. Both models have been validated through comparison of EMG measurements to simulated muscle activation [3,4] and inverse glenohumeral implant loading [7,8], but there has not been a consistent review of the modelled forces acting on the clavicle. One study does report large differences between measured and calculated forces when comparing the DSEM and cadaver measurements of the forces across the middle of the clavicle [8]. Neither model is focused on the clavicle, rather simulating the broader mechanics of the arm [3,4]. While DUEM has been implemented in a clavicle-centered model, adding some previously missing muscles, the joint definitions remained unchanged [7]. The movement of the clavicle is defined through the scapulohumeral rhythm [3,6,7], but how the clavicle is activated to conform to this rhythm differs: the AnyBody model uses both ligaments and muscles attached to the clavicle [4,5], while DUEM uses muscles and coordinate constraints [7]. The implementation of muscle attachments differs as deltoid arms are redefined in the AnyBody model [5] compared to the standards used in the DUEM model where, in turn, the trapezius, sternocleidomastoid and subclavius muscles are omitted [6]. The trapezius and sternocleidomastoid were added in the later clavicle-focused model [7]. Similarly, axial rotation of the clavicle is limited in both models, but the influence of this limitation is not further examined [5,6]. Conclusion These models offer a valuable framework, but a sensitivity analysis of the effects of muscle, ligament, and joint definitions and simplifications on clavicle loading is missing. Comparison to cadaveric measurements can allow a clavicle-specific verification. Combined, this can lead to a reliable clavicle-centered musculoskeletal model to be applied for fixation plate personalization. Adaptive Ultrasound Scan-Line Selection using Temporal Diffusion Models Oisín Nolan, Tristan S. W. Stevens, Wessel L. van Nierop, Ben Luijten, Ruud J. G. van Sloun Abstract: Subsampling is commonly used in medical imaging to minimize costs associated with data acquisition, such as acquisition time and dose. Recently, diffusion models have enabled learning complex image priors to create better reconstructions from partial observations than traditional methods. While a rich body of work exists for other medical imaging modalities, such as MRI and CT, using generative models to boost the efficiency of ultrasound acquisitions remains relatively unexplored. In this work, we present an agent-based model whose task is to reconstruct a sequence of ultrasound frames given a budget of k focused scan lines to observe per frame. The agent leverages a diffusion model to make intelligent decisions about which lines it should scan in order to generate accurate reconstructions. We train a diffusion model on sequences of 4 frames of ultrasound data from the CAMUS dataset to learn a spatiotemporal generative prior. At each frame, the agent acquires a subset of scanned lines and uses Diffusion Posterior Sampling to generate samples from the posterior distribution over full reconstructions. The agent iterates through the sequence of frames in this fashion, choosing which lines to scan to maximize expected information gain, approximated from the pixel-level posterior variance. The posterior sample with highest likelihood for each frame is used to produce a final reconstruction which is compared with the ground truth via mean absolute error for evaluation. The results show that the model can successfully reconstruct accurate images preserving important anatomical details using only 12.5% of the measurements per frame. We compare the maximum information sampling strategy to random sampling and find that maximum information sampling almost always outperforms random sampling. We also find that maximum information sampling with 4 lines outperforms random sampling with 6 lines, indicating that intelligent sampling, in this case, is worth 50% more measurements. Mutation-based Data Augmentation for Medical Imaging Segmentation shizhe Cai Abstract: In medical imaging segmentation, development of effective deep learning models and clinical deployment is significantly hampered by data scarcity and the high cost of labeling. This research addresses this challenge by proposing a novel data augmentation strategy. It minimizes the dependency on large datasets, and accelerates both development and deployment in clinical settings. Our goal is to maximize data diversity through data augmentation. It allows us to reduce the amount of medical imaging data required to train high-performance segmentation models. Specifically, we introduce a Mutation-based data generation method, which uses spatial transformations and deformations to mutate between object and background structures. These transformations enhance model learning by capturing variations in object location, shape, and intensity. Using only 10% of the original training data (39 samples) from the Medical Segmentation Decathlon Task01-BrainTumour dataset (MSD[1]), our method enables training of a simple, lightweight U-Net. It achieves a mean Dice Similarity Coefficient (DSC) of 0.612 on the test set, a result comparable to the state-of-the-art model (SwinUNETR[2]) performance of 0.644 mean DSC. This demonstrates that even with very limited data, our approach may yield high performance. Additionally, for datasets in the MSD collection with fewer than 20 samples, our method significantly enhances model performance. For example, on the MSD Task02-Heart dataset, this approach boosts the DSC by over 0.5, while on the Task04-Hippocampus dataset, it achieves an improvement of 0.3 DSC. Our findings challenge the conventional notion of data quantity in clinical model training, demonstrating that powerful augmentation can enable faster, cost-effective deployment of medical imaging segmentation models. [1]. Antonelli, Michela, et al. "The medical segmentation decathlon." Nature communications 13.1 (2022): 4128. [2]. Tang, Yucheng, et al. "Self-supervised pre-training of swin transformers for 3D medical image analysis." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2022. BOOST: Balanced Optimization of Occupational Satisfaction and Test performance Thom de Groot, Gijs van der Schot, Ajay Kottapalli, Claudine Lamoth, Charissa Roossien Abstract: Background: Chronic diseases account for approximately 75% of total deaths worldwide. Wearable sensors and point-of-care technology can play a significant role in preventing, monitoring, and screening people for these diseases in daily living and working environments. Modern "smart" office spaces are currently optimized to reduce operational costs. However, they often neglect workers' needs based on personal health and environmental factors. Advancements in technology and electronics have enabled the development of various types of wearable sensors that can collect data such as activity, light exposure and air quality. Sensors that can interact and communicate with each other is called a "Cyber Physical System". We aim to develop a setup that integrates technologies in a Cyber Physical Human System (CPHS). This innovative system will simultaneously monitor an individual's health and the surrounding environment. By integrating technologies that measure human health parameters such as activity, sleepiness and social interaction, and environmental conditions like air quality, noise and light, we can optimize the workplace for worker satisfaction and performance. The primary objective of this study is to create a CPHS capable of assessing multiple environmental factors alongside personal health metrics, ultimately enhancing the overall well-being of office workers. Method: This study will employ an A-B-A design. Each phase will take 2 weeks, and will include 10 to 20 healthy volunteers. Individual data between phases can be compared to see the differences. The first A-phase will be the baseline phase, during which data will be collected to establish a baseline. The collected data will relate to both environmental and human parameters. Environmental parameters include temperature, humidity, CO2 levels, light, and sound exposure. Human parameters encompass social interaction, stress levels, sleepiness, perceived fatigue and amount of physical activity. The B-phase will be an intervention phase where the user receives suggestions based on the measurements. These suggestions may include changing the intensity and/or colour of the desk light; putting on headphones or moving to a quieter place in the office (based on sound data); going for a walk or standing up for a while (based on movement data); and/or opening a door or window (based on air quality data). Feedback can be automated or delivered as a notification, depending on the suggestion. The second A-phase follows the intervention phase and has no intervention—This phase assesses whether the intervention has led to lasting effects on subject behaviour or if it reverts to the baseline established in the first phase. Assessing the Accuracy of Ligament Attachment Sites for Improved Slope-Correcting High Tibial Osteotomy Planning in ACL-Deficient Patients using Personalized Musculoskeletal Models Myrna de Wilde, Quinten Veerman, Periklis Tzanetis, Marnick Los, Nico Verdonschot Abstract: Slope-correcting high tibial osteotomy (HTO) redistributes load-bearing forces to prevent re-rupture of an anterior cruciate ligament (ACL) reconstruction (1). Current HTO planning relies on static images, leaving the impact on knee joint mechanics during dynamic activities underexplored. Personalized musculoskeletal models can address this, requiring precise modelling of knee ligamentous structures for valid simulations. This study examined the accuracy of morphed ligament attachment sites derived from the Twente Lower Extremity Model (TLEM) 2.0 dataset (2), comparing them with MRI-detected ground-truth attachment sites. Through this comparison, the work aimed at assessing whether morphed attachment points reliably model knee mechanics in subject-specific contexts. For two ACL-deficient subjects, attachment sites of the posterior cruciate ligament (PCL), lateral collateral ligament (LCL), and medial collateral ligament (MCL) were identified on MRI and mapped to a registered CT. The TLEM 2.0 template bone model, including ligament attachment sites, was personalised by morphing the template morphology to the subject-specific CT bone segmentations. The centroid of the MRI-based and morphed sites were compared using Euclidean distance measurements to assess distance errors in ligament attachment points. Findings revealed large variability, particularly in the tibial MCL attachment points, which had deviations between MRI-based and morphed points ranging from 11.10 to 18.98 mm across the two subjects. The variability for the other ligament attachment points were lower, ranging from 1.83 to 11.94 mm. Location differences in the two subjects showed consistent deviations, especially for the LCL on the femoral side, as well as for the MCL on both femoral and tibial side. This study revealed large variability in ligament attachment sites between MRI-based and morphed centroidal ligament points, suggesting uncertainties regarding the accuracy of the morphed points. These findings highlight the need to refine the morphing process. It should be noted that locating the attachment sites on MRI proved challenging due to limited through-plane resolution in the medial-lateral direction; reducing the slice thickness may improve accuracy. Future research should conduct a sensitivity analysis to assess the impact of ligament attachment errors on knee mechanics, focusing on ACL forces. Such advancements are vital for developing reliable subject-specific musculoskeletal models, enhancing HTO surgery outcomes. References 1. Nazzal EM, Zsidai B, Pujol O, Kaarre J, Curley AJ, Musahl V. Considerations of the Posterior Tibial Slope in Anterior Cruciate Ligament Reconstruction: a Scoping Review. Curr Rev Musculoskelet Med [Internet]. 2022 Aug 1 [cited 2024 Oct 10];15(4):291–9. Available from: https://link.springer.com/article/10.1007/s12178-022-09767-2 2. Carbone V, Fluit R, Pellikaan P, van der Krogt MM, Janssen D, Damsgaard M, et al. TLEM 2.0 – A comprehensive musculoskeletal geometry dataset for subject-specific modeling of lower extremity. J Biomech. 2015 Mar 18;48(5):734–41. Automated workflow for 3D printed, patient-specific NIV masks: Addressing fit and comfort challenges in chronic respiratory care Rens Wientjes, Casper Sandkuyl, Chien Nguyen, Michael Gaytant, Joris Jaspers Abstract: In the Netherlands, about 4,070 patients - including 250 children - are chronically dependent on non-invasive ventilation (NIV) due to conditions such as neuromuscular diseases, thoracic wall abnormalities, respiratory diseases and severe sleep apnoea. However, in up to 30% of cases, NIV masks fail due to poor fit, leading to air leakage, annoying noises, unnecessary alarms, inadequate ventilation and injury due to pressure on the face and, in children, can lead to hypoplasia of the midface. To address these problems, the University Medical Centre Utrecht (UMCU) has developed a workflow for creating 3D-printed, patient-specific NIV masks based on facial scans. This customised approach aims to improve mask fit, minimise air leakage and increase comfort, especially for the benefit of patients with unique facial anatomy. In the initial studies, a manual workflow was developed and printing techniques for flexible materials were selected. The face was scanned with a Revopoint Miraco 3D scanner. Mask models were 3D-printed with BASF Ultracur3D FL 300 resin on an Anycubic Photon Mono X 6Ks printer. This research aims to improve this workflow by automating the design of patient-specific masks using Python scripting for modelling masks based on facial scans and comparing the performance of these patient-specific masks with that of commercial masks. Performance metrics included air leakage, comfort and erythema. The accuracy of scanning the face in both seated and supine positions was also evaluated. Preliminary findings indicate high scanning accuracy, sufficient for the design of personalised NIV masks. The semi-automated model design script enabled fast and reproducible mask production, although commercial masks still outperformed custom-made masks in all areas tested. The findings highlight the feasibility of 3D-printed, patient-specific NIV masks while pointing out areas for improvement in fit, comfort and fully automated mask design, paving the way for personalised NIV masks. The influence of uncertainties in head tissue electrical conductivities on targeting metrics in transcranial temporal interference stimulation Paria Mansourinezhad, Debby Klooster, Rob Mestrom, Maarten Paulides Abstract: Transcranial temporal interference stimulation (tTIS) offers a non-invasive approach to stimulate deep regions in the brain with improved focality and steerability compared to other transcranial electrical stimulation (tES) methods. Using two or more pairs of electrodes, tTIS delivers high-frequency sinusoidal currents that differ slightly in frequency. When generated electric fields interfere spatially and temporally within the brain, they produce an amplitude-modulated electric field capable of coupling into neurons. The maximum envelope amplitude of the tTIS electric field can be localized to the target region by adjusting the electrode placement and the relative strengths of the currents applied with electrode pairs. Optimizing the stimulation montage to focus the electric field on a target region requires computational modeling, often based on the finite element method, to simulate electric fields, which depend on the electrical conductivity values of tissues in the head. However, accurately predicting these fields is challenging due to variability in tissue conductivity values across studies, which stems from differences in measurement techniques, experimental setups, and individual subject characteristics. In this study, we performed Monte Carlo simulations to systematically evaluate how uncertainties in head tissue conductivities influence optimized electric field metrics, such as target activation and off-target exposure. Using a high-resolution anatomical brain model (MIDA), we explored the effects of conductivity uncertainty on temporal interference fields targeting the primary motor cortex (a cortical region) and the left putamen (a deeper region). Given that no absolute threshold for the envelope amplitude of tTIS has been established in the literature for modulating neural activity, we analyzed uncertainties for a threshold range of 0.1 to 0.9 V/m. Our findings indicate that realistic conductivity variations lead to a wide distribution in targeting metrics. Furthermore, variations in threshold values substantially influence both target activation and off-target exposure, highlighting the sensitivity of these metrics to threshold selection. This analysis highlights the critical need to account for uncertainties in optimization metrics when designing stimulation montages for specific targets. Perspectives on Clinical Validation of Speech Biomarkers Mihir Kapadia, Robert-Jan Doll Abstract: Speech biomarkers have gained significant attention in recent years for their potential in clinical trials. Clinical applications of speech biomarkers include detecting and characterizing various neurological and psychological pathologies. Speech biomarkers can be acoustic – which provide a signal processing viewpoint, and linguistic – which analyse the structure and semantics of language in speech. Literature on clinical use of acoustic biomarkers focusses on spectrogram-based and mel-frequency spectral features into developing better models aimed at increasing accuracy of classification of disorders [1][2]. For linguistic biomarkers, the focus is on developing novel markers that reflect the changes in language syntax and structure, semantics and pragmatics in relation to specific pathologies [3][4]. The development of natural language processing models has led to state-of-the-art models and biomarkers for therapeutic areas such as identification of probable onset of Alzheimer’s disease and detection of speech aphasia in participants. While the need for a validation framework is recognized [5][6], limited attention is paid to discuss the aspects of clinical validation methods, feasibility and usability in clinical trials. The purpose of this study is twofold. First, we provide an overview of literature addressing clinical validation and feasibility of speech biomarkers. Second, an exploratory study involving a battery of neurophysiological and neuropsychological tests (NeuroCart [7]), as well as a speech collection task is being performed. Speech samples of 16 healthy participants will be used for the clinical validation of speech biomarkers. Speech samples are collected using a standardized protocol and task design; acoustic and linguistic biomarkers are derived and assessed for their repeatability, sensitivity and intra-class coefficients. In addition, the correlations of these biomarkers with NeuroCart endpoints are evaluated. This study attempts to bridge the gap of clinical validation of speech biomarkers by providing an assessment of correlations with other standardized test batteries. This study would provide the basis for pharmacodynamic relationships of speech biomarkers in future clinical trials and early-phase drug development. [1] G. Fagherazzi, A. Fischer, M. Ismael, and V. Despotovic, “Voice for Health: The Use of Vocal Biomarkers from Research to Clinical Practice,” Digit. Biomarkers, vol. 5, no. 1, pp. 78–88, 2021, doi: 10.1159/000515346. [2] A. Tsanas, M. A. Little, P. E. McSharry, and L. O. Ramig, “Nonlinear speech analysis algorithms mapped to a standard metric achieve clinically useful quantification of average Parkinson’s disease symptom severity,” J. R. Soc. Interface, vol. 8, no. 59, pp. 842–855, 2011, doi: 10.1098/rsif.2010.0456. [3] V. Boschi, E. Catricalà, M. Consonni, C. Chesi, A. Moro, and S. F. Cappa, “Connected speech in neurodegenerative language disorders: A review,” Front. Psychol., vol. 8, no. MAR, 2017, doi: 10.3389/fpsyg.2017.00269. [4] N. Nevler, S. Ash, D. J. Irwin, M. Liberman, and M. Grossman, “Validated automatic speech biomarkers in primary progressive aphasia,” Ann. Clin. Transl. Neurol., vol. 6, no. 1, pp. 4–14, 2019, doi: 10.1002/acn3.653. [5] V. Berisha and J. M. Liss, “Responsible development of clinical speech AI: Bridging the gap between clinical research and technology,” npj Digit. Med., vol. 7, no. 1, pp. 1–12, 2024, doi: 10.1038/s41746-024-01199-1. [6] J. Robin, J. E. Harrison, L. D. Kaufman, F. Rudzicz, W. Simpson, and M. Yancheva, “Evaluation of Speech-Based Digital Biomarkers: Review and Recommendations,” Digit. Biomarkers, vol. 4, no. 3, pp. 99–108, 2020, doi: 10.1159/000510820. [7] Measuring a wide range of CNS effects in a pharmacological context NeuroCart : the next generation of multifunction , standardised CNS test batteries Ultrasound-based tracking of abdominal aortic aneurysm growth: Insights on strain, curvature and thrombus Mirunalini Thirugnanasambandam, Esther Maas, Arjet Nievergeld, Marc van Sambeek, Richard Lopata Abstract: Estimation of rupture risk in abdominal aortic aneurysms (AAA) has gathered significant attention within the scientific community. However, it is crucial to acknowledge that this critical event results from the underlying growth and remodeling processes. To understand the nature of these processes, it is important to identify the relevant image-based biomarkers and characterize their development with patient-specific precision. In this study, co-evolution of strain, curvature and thrombus is analysed using ultrasound (US) images. 4D-US images were acquired from thirty AAA patients over atleast three follow-ups in a supine position at 4-8 Hz. Image volumes were segmented at diastole, and a 3D speckle tracking algorithm was used to track the AAA wall to systole. Strain maps corresponding to the AAA wall displacement were generated using 3D least square-based strain estimation. In addition to the maximum principal, radial, and circumferential strains, the corresponding principal strain directions were also evaluated. Patient-specific geometries of the wall were used to generate the corresponding spatial distribution of Gaussian and Mean curvature. Visibility of the ILT in these patients allowed lumen segmentation, enabling evaluation of the spatial variation of ILT thickness. AAA geometries from the same patient were co-registered by matching their centerlines before temporal analyses. Structural similarity index maps (SSIM) were used to evaluate spatiotemporal changes in the parameters. Regions of increased ILT thickness typically corresponded to low strain areas, indicating an increase in wall stiffness behind ILT. Global SSIM assessment showed that radial strain and principal strain directions changed significantly over time. However, circumferential strain changes were similar to those in principal strain. Gaussian and mean curvature values did not show any sharp changes. In general, when an ILT was present, the increase in ILT thickness was much faster compared to changes observed in any other parameter. In the future, evolution of stress and other hemodynamic parameters of interest will also be studied alongside these markers. Development of the PsyCart: a novel RDoC based emotional test battery Soma Makai-Bölöni, Laura Borghans, Ingrid van den Heuvel, Annika de Goede, Gabriel Jacobs, Robert-Jan Doll Abstract: An emotional test battery can aid in understanding compounds' underlying mechanisms of action and in predicting the efficacy of novel psychiatric drugs in their early development phase (Bland et al., 2016a). For example, the P1vital® (Murphy et al., 2008) has been used to predict effects of antidepressant drugs and has been validated in healthy volunteers and in patients with depression (Harmer et al., 2003, 2009, 2011). At the Centre for Human Drug Research (CHDR), a test battery, named the NeuroCart® test battery, has already been used to measure various aspects of cognition and neurophysiology. However, a test battery that assesses emotional processing is currently lacking. The Research Domain Criteria (RDoC) was used as the theoretical framework (Cuthbert, 2022) to understand which emotional domains to investigate. Here, we focus on the Grip Effort Task (GET) and Facial Emotion Recognition (FERT) tasks. These were implemented to assess reward responsiveness and social communication, respectively. However, before these tasks can be utilized in clinical trials they must show repeatable and consistent results to assess the effects the effects of pharmacological interventions. For the present feasibility study, 16 healthy volunteers were enrolled at the CHDR (Leiden, the Netherlands). The GET and FERT were assessed multiple times. Study endpoints were computed, and the repeatability, potential sensitivity, as well as the feasibility were assessed. The results of this experiment will be discussed, and directions for future research will be outlined.   Forward dynamic optimization of the shoulder girdle motion Frans C.T. van der Helm, H.E.J. (DirkJan) Veeger, Thomas Geijtenbeek Abstract: Problem Statement: A wide variety of upper extremity models exist, though not all of them have been thoroughly validated. Most important items are the number of muscle lines of action, implementation of clavicular ligaments, glenohumeral joint restraints and scapulothoracic contact. Shoulder girdle motions are usually subconscious motions subserving motions of the arm. In the so-called scapulohumeral rhythm motions of the scapula are described as a function of the humeral motions. Models can be validated by comparison of the prediction of scapular motions during arm motion with actually recorded scapular motions. Goal: The goal of this research is to demonstrate a novel way to validate biomechanical models of the upper extremity, including the model parameters and optimization criteria, using predictive simulations of the subconscious motions of the scapula. Method: A biomechanical model of the upper extremity has been developed for Scone (Geijtenbeek, 2019). Scone is open source software (http://scone.software), accelerated by Hyfydy (Geijtenbeek, 2021). Model parameters of a simplified biomechanical model has been adopted from OpenSim (Seth et al., 2016). The model simplication is partly in the reduction of number of muscle lines of action and in the kinematic structure of the shoulder girdle, i.e. representing the restraint of the scapulathoracic contact. In Hyfydy the joint restraints are represented as force – displacement structures, which accelerates the simulation of motion equations considerably, allowing for predictive simulations with task-based optimization objectives. In addition, this representation seamlessly incorporates the scapulothoracic restraints. Scapular motions have been recorded in an extensive study (Nikooyan, 2010), in which also EMG and glenohumeral (GH) joint reaction forces were recorded with an instrumented prosthesis. Model predictions of the scapular motions have been compared to these measurements. Results and conclusions: In a detailed analysis of a variation of model structures and model parameters, the validity of each combination has been compared. It is concluded that reduction of the number of muscle lines of action results in peak forces in the remaining muscle lines of action. The representation of the scapulathoracic contact as force restraints is a proper way to model the physical contact. Presumably as a result of the stability requirements for the glenohumeral joint, the GH joint reaction force is increasing above 90 degrees elevation, similar as recorded with the instrumented endoprosthesis.

end %-->