13:00   Vascular - II
Inter-donor variability in the tensile and compressive properties of in vitro human thrombi
Rachel Cahalane, Ali Akyildiz, Frank Gijsen
Abstract: Introduction The efficacy of endovascular treatment of acute ischemic stroke has been demonstrated in large clinical trials. However, substantial reperfusion is not achieved in up to 56% of cases [1]. The response to treatment is thought to be significantly influenced by the thrombus mechanics [2]. While thrombus composition is linked with mechanics [3], composition alone cannot explain the variation in compressive stiffness [4]. Previous data suggest that inter-donor variations exist [5], but this has not been investigated. Further, very limited studies examine thrombus analogs produced from human blood [3]. Here, we report on the tensile and compressive behaviour of thrombus analogs made from the blood of different human donors in a variety of compositions. Materials and Methods Fresh blood was obtained from 6 healthy human volunteers (3 male, 3 female). Blood cell counts and fibrinogen levels were quantified. Nine different compositions were made: whole blood, platelet-poor plasma, platelet-rich plasma, 5%, 10%, 20%, 40%, 60%, 80% RBC clots. The stiffness of the thrombi were determined by performing unconfined compression and uniaxial extension testing. The composition of the thrombi was confirmed with histology. Results RBC volumetric ratios do not produce thrombi with equivalent RBC ratios (5% volumetric ratio produces a thrombus with >70% RBC content). The stress-strain response of thrombi under compression is non-linear, especially at high strains (>40%). In contrast, the tensile stress-strain response is largely linear. There are large variations in the tensile and compressive stiffness values observed for whole blood clots from different donors (stiffness at compressive strains of 75% (E75%) range from 12 to 36 kPa). Donor age, platelet count and fibrinogen levels appear to explain some of this variation (rp = 0.79, 0.74 and 0.71, respectively). Conclusion Thrombi exhibit compression-tension asymmetry. Thrombi from different donors exhibit variable mechanical properties that can be partially explained by donor demographics and hemostatic measurements. It may be possible to predict inter-patient variation in thrombus response to treatment using a clinical mechanical assay. References [1] A. J. Yoo and T. Andersson, “Thrombectomy in acute ischemic stroke: Challenges to procedural success,” J. Stroke, vol. 19, no. 2, pp. 121–130, 2017, doi: 10.5853/jos.2017.00752. [2] S. Johnson, S. Duffy, G. Gunning, M. Gilvarry, J. P. McGarry, and P. E. McHugh, “Review of Mechanical Testing and Modelling of Thrombus Material for Vascular Implant and Device Design,” Annals of Biomedical Engineering, vol. 45, no. 11. pp. 2494–2508, 2017, doi: 10.1007/s10439-017-1906-5. [3] Cahalane et al., “A review on the association of thrombus composition with mechanical and radiological imaging characteristics in acute ischemic stroke,” J. Biomech., 2021. [4] N. Boodt et al., “Mechanical Characterization of Thrombi Retrieved with Endovascular Thrombectomy in Patients with Acute Ischemic Stroke,” Stroke, 2021, doi: 10.1161/STROKEAHA.120.033527. [5] W. Merritt et al., “Quantifying the mechanical and histological properties of thrombus analog made from human blood for the creation of synthetic thrombus for thrombectomy device testing,” J. Neurointerv. Surg., vol. 10, no. 12, pp. 1168–1173, 2018, doi: 10.1136/neurintsurg-2017-013675.
Towards a fully automatic workflow to assess wall-stress analysis of abdominal aortic aneurysms including intraluminal thrombus using 4D-ultrasound
Arjet Nievergeld, Judith Fonken, Esther Maas, Mirunalini Thirugnanasambandam, Frans van de Vosse, Marc van Sambeek, Richard Lopata
Abstract: An abdominal aortic aneurysm (AAA) is a localized dilation 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 a sub-optimal criterion for rupture risk prediction. Biomechanical models can improve rupture risk prediction using patient-specific geometries obtained from imaging modalities such as CT and ultrasound (US). US is safer than CT, especially when follow-up studies are needed. Moreover, time-resolved, 3-dimensional US imaging adds temporal information, required for mechanical characterization (elastography), which is important for personalized risk prediction. It is hypothesized that intraluminal thrombus (ILT) lowers the wall stress and therefore should be included in rupture risk assessment. Therefore, a 4D US-based workflow was designed, including automatic segmentation, meshing and modelling algorithms to perform Finite Element Analysis (FEA) on patient-specific AAA geometries including ILT. The geometry of the AAA including the lumen, thrombus, and vessel wall was determined from 4D US images using a star-Kalman and 3D-snake algorithm, assuming a uniform wall thickness of 2 mm. The automatic segmentation was validated using CT and manual US-based segmentations. The segmentation was successful in 11 out of 37 patients, which was comparable to the manual segmentations. The performance of the segmentation, both manual and automatic, was strongly dependent on the visibility of the thrombus. The vessel wall was meshed using quadratic hexahedrons and the thrombus was meshed using pyramids and quadratic tetrahedrons created with TetGen and the Gibbon Matlab toolbox. First results show a decrease in wall stress after including the thrombus, which is as hypothesized. To our best knowledge no FEA of AAAs including ILT and based on 4D US has been reported. First results are promising, and could be further improved in future research. The performance of the segmentation can be improved using Doppler or contrast-enhanced US. Besides, this model can be expanded with a stiffness estimation algorithm, combining speckle tracking and blood pressure. Eventually, this fully automatic workflow can be used in a longitudinal study to estimate the patient-specific AAA rupture risk in a more accurate way.
Local characterization of atherosclerotic plaque rupture through tensile testing. Digital image correlation and nano-indentation
Su Guvenir Torun, Pablo de Miguel Munoz, Ali Cagdas Akyildiz
Abstract: Biomechanical analysis of the atherosclerotic plaque rupture is crucial since these events occur when the local stresses in the plaque tissue exceed the local tissue strength. Currently, the knowledge on the tissue’s mechanical properties is limited to average properties, obtained by homogeneity assumption and global deformation measurements in ex-vivo mechanical tests [1]. However, the local plaque tissue mechanical properties are essential for rupture risk assessment, due to the heterogeneity of the plaque tissue. In this study we aim to capture local properties (stiffness and failure) of atherosclerotic human carotid plaque tissue by using uniaxial tensile testing, digital image correlation (DIC) and the nano-indentation technique. For the preliminary work, three rectangular shaped tissue strips were prepared from an atherosclerotic human endarterectomy sample, and a speckle pattern was applied to the lumen side. The speckled tissue strips were tested by uniaxial tensile testing, until the complete rupture. Then, the DIC analysis was performed for local strain measurements until rupture initiation. In order to investigate a possible correlation with the local tissue stiffness and the tissue strength, the local mechanical characterization was performed with the nano-indentation technique. After analyses, the highest stiffness locations were observed close to rupture sites. 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 and/or high shear strain regions were correlated with the rupture areas. This correlation could indicate different failure mechanisms due to the underlying microstructural organization. In this work, we conducted a preliminary study for our new pipeline, that can provide local failure characteristics of highly heterogenous plaque tissue, which is more relevant for real-life plaque rupture than the conventional average ultimate tensile properties obtained by tissue homogeneity assumption. We believe that a possible correlation between the local mechanical and structural tissue properties will provide great insights for the failure mechanisms involved in plaque rupture. Hence, in future work, the new pipeline will be applied on approximately 20 human carotid endarterectomy samples, and will be advanced by obtaining the microstructural (collagen) information with second harmonic imaging. [1] Akyildiz A C et al., J Biomech, 47.4:773-783, 2014.
Image registration via deep learning towards image-guided interventions
Zhen Li, Maria Elisabetta Mancini, Giovanni Monizzi, Daniele Andreini, Giancarlo Ferrigno, Jenny Dankelman, Elena De Momi
Abstract: Clinical aim: Cardiologists highlight the need for an intra-operative 3D visualization to assist interventions. The intra-operative 2D X-ray/Digital Subtraction Angiography (DSA) images in the standard clinical workflow lack depth perception. State-of-the-art: Compared with image-to-image registration, model-to-image registration is an essential approach taking advantage of the reuse of pre-operative 3D models reconstructed from Computed Tomography Angiography (CTA) images. Traditional optimized-based registration methods suffer severely from high computational complexity. Moreover, the consequence of lacking ground truth for learning-based registration approaches should not be neglected. Method: To overcome these challenges, we introduce a model-to-image registration framework via deep learning for image-guided endovascular catheterization. In order to find the correspondence between a pre-operative model and intra-operative images, this framework firstly performs autonomous vessel segmentation from fluoroscopy images. A deep residual U-net architecture is employed, thanks to its fast learning convergence and efficient spatial information propagation without degradation. After that, a model-to-image matching via Convolutional Neural Network (CNN) is introduced. For this study, image data were collected from 10 patients who performed Transcatheter Aortic Valve Implantation (TAVI) procedures. Results: It was found that vessel segmentation of test data results in median values of Dice Similarity Coefficient, Precision, and Recall of 0.75, 0.58, 0.67 for femoral artery, and 0.71, 0.56, 0.74 for aortic root. The automatically segmented network behaves better than the manual annotated one, and it recognizes part of vessels that were not labelled manually. Image matching between the transformed moving image and the fixed image results in a median value of Recall of 0.90. Conclusions: The proposed approach achieves a good accuracy of vessel segmentation and a good recall value of model-to-image matching. The proposed framework is based on the reuse of the pre-operative model that is reconstructed from CTA images for diagnosing and size measurement, and there is no interference with the standard clinical workflow. Future work: Future improvements include integrating augmented reality, performing user-end evaluation in the operating room, and extending to a deformable registration approach considering the vessel deformations due to the device contact during the procedure.
Shear stress related plaque progression of LIPID rich plaques in human coronary arteries
Eline Hartman, Giuseppe De Nisco, Suze-Anne Korteland, Frank Gijsen, Anton van der Steen, Joost Daemen, Jolanda Wentzel
Abstract: Introduction Wall shear stress (WSS) of the blood at the vessel wall plays a crucial role in atherosclerotic plaque development. Low WSS is known to be responsible for local regions of endothelial dysfunction and these are the locations notorius of lipids infiltration into the vessel wall resulting in atherosclerotic plaques. Moreover, lipids effect the inflammatory cascade and stimulate plaque progression as well. The aim of this study was to investigate the mutual interaction between WSS and intraplaque lipids on plaque progression in human coronary arteries. METHODS 54 Patients presenting with an acute coronary syndrome (ACS) underwent NIRS-IVUS and OCT assessment of a non-culprit artery at baseline and 12-month follow-up. Subsequently after 1 month CT angiography was performed. Vessel-specific flow was measured using Doppler wire and computational fluid dynamics was performed to calculate WSS in 3D reconstructions based fusion of CT and intravascular imaging. For each vessel, 1.5-mm segments were identified, matched and divided into 45 sectors. IVUS-derived percentage atheroma volume (PAV), and NIRS and OCT-derived plaque components were projected on the 3D-reconstruction. Plaque growth based on PAV change was assessed comparing lipid-rich plaques, non-lipid rich plaques and plaque free wall segments of both OCT as well as NIRS derived plaque components. RESULTS Both NIRS- and OCT-detected lipid-rich sectors showed significant higher plaque progression than non-lipid rich plaques or plaque free regions (see figure). Exposure to low WSS showed significant higher plaque progression than exposure to mid or high WSS, even in regions classified as plaque free wall. Furthermore, low shear stress and the presence of lipids had a synergistic effect on plaque growth resulting in the highest plaque progression in lipid-rich regions exposed to low shear stress. CONCLUSION This study demonstrates that lipid-rich regions, both detected by NIRS as well as OCT exposed to low WSS presents with the highest plaque growth after one-year of follow-up. The presence of lipids and low WSS proved to have a synergistic effect on plaque growth.
Carotid artery segmentation in ultrasound images using an unsupervised domain adaptation method
Luuk van Knippenberg, Joerik de Ruijter, Arthur Bouwman, Richard Lopata, Ruud van Sloun, Massimo Mischi
Abstract: Background Hemodynamic monitoring is used in critically ill patients to assess the cardiovascular system. The common carotid artery is easily accessible by ultrasound and may therefore be suitable for continuous flow monitoring. Previously, we showed that cross-sectional Doppler is a robust and more operator-independent alternative to longitudinal flow measurements [1]. By assuming a cylindrical vessel geometry, the intersection with the ultrasound plane will result in an ellipse, of which the parameters can be used to estimate the Doppler angle and vessel diameter. However, for accurate flow estimates the segmentation method should be robust, fast and automatic, which is challenging using conventional signal processing methods. Deep learning models have shown success in a wide variety of image segmentation tasks, but traditionally require extensive labeled datasets to train, which are often unavailable for in-vivo data. Methods In this work, we demonstrate an unsupervised domain adaptation method that allows us to apply to in-vivo data a model originally trained on mostly simulated data. By using prior knowledge on the elliptical geometry of the predictions, unexpected outputs can be identified via fitting mismatch and corrected for by proper adaptation of the weights. First, the model is trained on a labeled dataset consisting of 2000 images, of which 95% is simulated data and 5% is data acquired on phantoms and volunteers using an EPIQ 7G ultrasound scanner. Then, the model is further trained on an unlabeled dataset consisting of 2000 in-vivo images acquired using a clinical MyLabOne Vascular Ultrasound system [2]. Labels are generated by iteratively applying the latest weights and fitting an ellipse to the largest connected region in the prediction. Performance metrics are computed using manually labeled images. Results/Discussion The mean Dice similarity coefficient increased from 0.86±0.26 to 0.94±0.07 by domain adaptation, which is slightly better than the Dice similarity coefficient obtained using the Star-Kalman algorithm (0.92) designed for this specific task [2]. Hence, the model successfully adapted from simulations to in-vivo data without human supervision or fine-tuning. These results suggest that prior information on the shape of the mask can be used to keep training on unlabeled data, such that no manual annotations are required.

end %-->