Paper Submission & Registration
8th Dutch Bio-Medical Engineering Conference
13:40   Vascular - I
Chair: Jolanda Wentzel
15 mins
Heterogeneous material property characterization of atherosclerotic human carotid arteries
Su Guvenir, Hakki M, Torun, Hendrik H.G. Hansen, Ali C. Akyildiz
Abstract: Stroke, a leading cause of mortality worldwide, is commonly initiated by the biomechanical event of plaque rupture in atherosclerotic human carotid arteries. The current clinical tools lack the predictive power for the risk assessment, and there is a need for novel tools with a biomechanical criterion for rupture prediction. It is well established that high structural stresses are correlated with the rupture events [1], and their local assessment is of great importance. Finite element models are used to obtain plaque tissue stresses, and they require accurate representation of the local, heterogeneous material properties. However, this information is currently provided as homogenous and average material properties obtained by mechanical tests that do not closely mimic physiological loading conditions [2]. In this work, we addressed the mentioned limitations by building a pipeline comprising of ex-vivo mechanical inflation testing, up to physiological pressure levels, combined with a high-resolution 21MHz ultrasound imaging, high field 7 Tesla magnetic resonance imaging based two-dimensional finite element modelling, image registration, and a sample efficient Deep Partitioning Tree based Bayesian Optimization. Our pipeline, namely the inverse finite element modelling approach, was used to characterize 5 atherosclerotic human carotid arteries. 13 cross sections were selected based on their plaque density and location, and were characterized successfully. 11 of these cases achieved convergence in less than 10% normalized mean square error. Moreover, by inverting the model trained in the optimization protocol, we were able to check the uniqueness of the solutions. In this study, we characterized heterogeneous intima and wall material properties of atherosclerotic human carotid arteries by building a pipeline involving the inflation testing that conserves the intact structure of the tested arteries and mimics the physiological in-vivo-like loading conditions. Moreover, the sample efficient optimization algorithm allowed us to predict multiple parameters converged around the global optimum in a cost-effective manner. Thus, the introduced pipeline has the great potential for the clinical translation. The clinical magnetic resonance imaging could be used to obtain the accurate morphological information of the carotid arteries, whereas the ultrasound system could be used to detect the deformation as we have used in our ex-vivo pipeline.
15 mins
Geometry and strain assessment of the healthy and aneurysmal abdominal aorta with a semi-3D ultrasound imaging approach
Larissa Jansen, Hans-Martin Schwab, Frans van de Vosse, Marc van Sambeek, Richard Lopata
Abstract: Risk stratification of abdominal aortic aneurysm (AAA) patients is based on monitoring the aneurysm diameter using 2D ultrasound (US) imaging. This criterion is not adequate for every case, leading to fatal haemorrhages or surgical overtreatments. 4D US imaging combined with mechanical modeling is a promising alternative for risk assessment. With this tool, the geometry and mechanical state of the aneurysm can be assessed. However, the limited field of view (FOV) and poor spatial resolution of 4D US imaging impede segmentation quality. Furthermore, low temporal resolution cumbers strain estimation. Therefore, we propose to perform a free-hand sweep with a tracked 2D US probe to achieve a large FOV combined with local strain data. The proposed method was tested on ten healthy volunteers and fourteen AAA patients. Computed Tomography (CT) images were available for five patients serving as geometry ground truth. B-mode images and probe orientation data (Curefab technologies, DE) were obtained while moving a curved array probe (Esaote, NL) free-hand on the abdomen. Next, at five locations radio frequency (RF) data was collected. The geometry was extracted by first indicating the aorta centre in the outer frames. Next, the wall-lumen contours were detected automatically in each frame using a Star-Kalman algorithm. Finally, the heart frequency was detected and used to filter out the pulsatile motion of the contours. An iterative 2D block-matching algorithm was applied to the RF-data to compute local strain data. CT-based geometries were obtained semi-automatically using Hemodyn (Philips and Eindhoven University of Technology, NL). The US-based geometries were registered with these geometries using an iterative closest point algorithm and were compared by computing the Similarity Index (SI) and Hausdorff distance (HD). Geometries obtained with the proposed semi-3D approach showed good similarity with CT geometries, i.e., an overall median SI and HD with interquartile range of 0.90 (0.89 – 0.91) and 5.1 (4.1-6.4) mm. The best segmentation quality was achieved in upper and lower wall regions. Moreover, a major part of the AAA (up to 10cm in length) can be captured together with local strain information. Ultrafast and multi-perspective US imaging are investigated to further improve functional imaging.
15 mins
Towards including the intraluminal thrombus in patient-specific models of AAAS based on 4-D ultrasound
Arjet Nievergeld, Esther Maas, Joerik de Ruijter, Frans van de Vosse, Marc van Sambeek, Richard Lopata
Abstract: An abdominal aortic aneurysm (AAA) is a localized dilatation of the aorta, which in case of rupture has a mortality rate of 80%. Current clinical guidelines of intervention are based on AAA diameter, which has been proven to be an inadequate criterion. Biomechanical models can improve the prediction of rupture risk in a more patient-specific way, using e.g. CT or ultrasound (US) imaging [1, 2]. US is safer compared to CT and adds temporal information for mechanical characterization of the AAA. It is hypothesized that intraluminal thrombus (ILT) lowers the peak wall stress [3]. Therefore, it is needed to characterize the ILT and include it in patient-specific models. US imaging of the ILT is challenging considering the low contrast. The objective of this study is to determine the patient-specific geometry of the AAA wall, ILT and lumen in 4-D US images. Methods 4-D US images of 16 AAA patients, known to have an ILT, are segmented using a fully automatic segmentation method. The inner vessel wall is segmented using a slice by slice star-Kalman algorithm [4]. The lumen-thrombus interface (LTI) is determined combining the Star-Kalman algorithm with a 3D-snake [5]. The resulting US-based geometry of the inner vessel wall and LTI are compared with CT-based geometries, calculating the similarity index (SI) and the median Hausdorff distance (HD) per slice [1,6]. Results The median SI for the vessel wall and LTI are 0.90 (0.83-0.95) and 0.84 (0.75-0.91), respectively. The median Hausdorff distance for the inner vessel wall and LTI are 4.5 (3.4 – 5.5) mm and 5.6 (3.0 – 7.5) mm. Discussion The first results show good agreement between the US-based geometry and the CT-based geometry for all patients. In future work, the segmented geometry will be used in combination with blood pressure and speckle tracking [1], to determine the mechanical properties of the thrombus and to create a patient-specific model of the AAA. This will allow to evaluate the effect of thrombus on the stress distribution in the aortic wall and to assess the patient-specific rupture risk in a more accurate way. References: 1. Van Disseldorp et al, Eur J Vasc Endovasc Surg, 59: 81-91, 2020 2. Kok et al, J Vasc Surg 61: 1175-1185, 2015 3. Inzoli et al, Eur J Vasc Surg, 7:667-674, 1993 4. Abolmaesumi et al, IEEE Trans Robot, 18: 11-23, 2002 5. Kass et al,. Int J Comput Vis, 1:321–331, 1988. 6. Rote et al, Inform Process Lett, 38: 123-127, 1991
15 mins
Robust Flow Monitoring Using Cross-sectional Doppler and Adaptive Transmits
Luuk van Knippenberg, Ruud van Sloun, Arthur Bouwman, Sergei Shulepov, Massimo Mischi
Abstract: Background Doppler ultrasound is the most common technique for non-invasive quantification of blood flow, which can be used to assess the cardiovascular condition. To estimate flow, the operator has to obtain a longitudinal image in which both Doppler angle and vessel diameter can be estimated. However, moving towards ultrasound-based monitoring, the need for an operator should be overcome. Previously, we showed that the Doppler angle can be estimated by fitting an ellipse to the vessel in cross-sectional ultrasound acquisitions, resulting in accurate angle-corrected velocities [1]. Yet, to achieve accurate estimates under various imaging conditions, thereby being operator-independent, the transmit steering angle should adaptively be optimized. Methods In this work, we demonstrate an implementation on a research ultrasound system (Verasonics) that adaptively shifts the spectral Doppler focus to the center of the vessel, varies the aperture width to realize a fixed F-number (1.5), and uses a steering angle that minimizes the Doppler angle, based on power Doppler segmentation. The IQ data is then processed using a moving gate, such that a velocity profile can be estimated from which the average velocity is computed. The Verasonics sequence was tested in simulation mode, where various vessel orientations and positions were simulated using a parabolic velocity profile (v_0=0.5 m/s). The effect of other transmit/receive parameters such as transmit frequency, pulse length, apodization and gate size were explored using response surface methodology [2] to optimize the results. Results/Discussion Our simulations show that the transmit focus, steering angle and aperture width can adaptively be changed based on vessel segmentation by power Doppler, enabling ultrasound-based flow monitoring. Compared to using non-steered beams, the estimated velocity profile and resulting average velocity are more accurate and have a smaller spread when adaptive beam steering is used (-1.9±10.5% and 5.6±14.4% error, respectively). The response surface analysis shows that the gate size and maximum steering angle are the most significant factors in determining the estimated average velocity, where a small gate size (0.5 mm) and maximum steering angle of 14 degrees are most optimal. In the future, we will also investigate the performance of the proposed method both in-vitro and in-vivo.

end %-->