Friday, December 1, 2023

CT-based comparison of porcine, ovine, and human pulmonary arterial morphometry – Scientific Reports

To compare the pulmonary artery morphometry of the two animal models with that of humans, retrospective computed tomography (CT) data available from previous animal studies as well as clinical routine was used. PAPS can usually be implanted in both the left and right PA, even though manufacturer recommendation might favor a dedicated implantation site. Spatial dimensions of current PAPS systems are 15.0 mm × 3.4 mm × 2.0 mm (CardioMEMS) and 19.3 mm × 3.8 mm × 1.9 mm height (Cordella) in length, width and height respectively. Target diameters of the implantation site for the CardioMEMS device are between 7 and 11 mm12 and between 12 and 26 mm for the Cordella system13, respectively. For fixation of the device within the pulmonary artery wires are attached to the sensor body. An additional, a novel device with sizes similar to those of the CardioMEMS and Cordelia systems will also be evaluated, for which the target diameters of the implantation site are estimated to lie between 9 and 14 mm as optimal. This defined the domain of interest for the anatomical study, which included the MPA, LPA, and RPA as well as all side branches of the LPA and RPA. 3D geometries of the PA were reconstructed from available CT image data aiming at reconstruction of the vessels until LPA and RPA diameters were below 8 mm or as long as the image data allowed.

Imaging data

Data of 41 domestic pigs (Swiss Large White, 3 castrated males, 38 females), weighing 82.6 ± 18.8 (range 52–117 kg) kg, and being 3–6 months old, that underwent pre-operative cardiac CT for a different project in the Department of Health Sciences and Technology, ETH Zürich, were re-used in this study. Acquisition was done by dual-source multi-slice spiral CT scanner SOMATOM Definition Flash (Siemens Healthineers, Forchheim, Germany) with 120 kV tube voltage as well as (0.391–0.684 mm) × (0.391–0.684 mm) in-plane resolution and (0.3 mm) slice thickness. These animal studies were approved by the local Committee for Experimental Animal Research (Cantonal Veterinary Office Zurich, Switzerland) under the approval numbers 152/2013, 219/2016 and 138/2017. More information about the experiment for which the CT information was acquired can be found in the respective studies11,14.

Additionally, pre-operative CT image data of 14 sheep (breeds: Ovis aries, German heath sheep; 7 males, 7 females) weighing 49.1 ± 6.9 (range 40 – 63 kg) kg, and being of 9.0 ± 0.8 months old, were available from a project aiming at development of a pulmonary valve prosthesis. Dual-source multi-slice spiral CT scanner SOMATOM Definition Flash (Siemens Healthcare GmbH, Erlangen, Germany) with 100 kV tube voltage, (0.666 – 0.977 mm) × (0.666–0.977 mm) in-plane resolution and (0.7 mm) slice thickness was used for data acquisitions. This study was approved by the State Office of Health and Social Affairs Berlin and conducted according to the Federation of Laboratory Animal Science Associations guidelines. More information about the CT imaging procedure can be found in the respective study14. Both animal experiments from which this data were derived were performed in accordance with all relevant guidelines and regulations. This study exclusively re-uses existing imaging data of animals that was acquired in earlier studies and were not related to this research at all.

Finally, retrospective CT image data of 49 aortic stenosis patients, weighing 76.8 ± 18.2 kg, with an age distribution of 81 ± 7.6 years (median 82.5 years and interquartile range (IQR) of [77.75–85.0]), and a female percentage of 60% were used for reconstruction of human pulmonary artery samples. CT data sets of the entire heart were acquired as part of transcatheter aortic valve implantation (TAVI) planning between February 2019 and October 2020 at our clinical centre. CT data were acquired with wide area-detector volume CT scanners: Aquilion One Vision (Canon Medical Systems, Tochigi, Japan) or Revolution CT (GE Healthcare, Chicago, IL, USA) with 100 kV tube voltage, (0.390–0.648 mm) × (0.390–0.648 mm) in-plane resolution and (0.5–1.0 mm) slice thickness. The use of retrospective data for this study was approved by the institutional review board (Ethikkommission Charité–Universitätsmedizin Berlin, approval number EA2/004/21). Individual informed consent was waived by the IRB due to the retrospective nature of this study. Research using this data was performed in accordance with all relevant guidelines and regulations. Additional information on the CT acquisition and the data set can be found in the respective study15.

Imaging procedures for both sheep and pigs were performed under general anaesthesia. Pigs were sedated using an intramuscular injection of ketamine (Ketasol®-100 ad; Dr. E. Graeub AG, Berne, Switzerland; 15 mg/kg body weight), azaperone (Stresnil® ad; Elanco Tiergesundheit AG, Basel, Switzerland; 2 mg/kg body weight) and atropine (Atropinsulfat KA vet 0.1%; Kantonsapotheke, Switzerland; 0.05 mg/kg body weight). Anesthesia was induced by an intravenous administration of propofol (Propofol ®- Lipuro 1%, B. Braun Medical AG; Sempach, Switzerland; 1–2 mg/kg body weight) to achieve relaxation and swallow-reflex diminishment sufficient for intubation. Anesthesia was maintained with propofol (5–10 mg/kg/h). Sheep were sedated using intravenous thiopental 7.5–10 mg/kg (Trapanal® 0.5 g, Inresa Arzneimittel GmbH, Germany) and propofol 1–2.5 mg/kg (Propofol-Lipuro 1%®, Braun, Melsungen, Germany). The sheep were intubated, and a gastric tube was placed into the rumen for gas evacuation. Total intravenous anesthesia was performed using propofol (2.5–8.0 mg/kg/h) and ketamine (2–5 mg/kg/h). All CT images were acquired before any further intervention, such as device implantation, and no further medication was administered.

Geometry reconstruction

CT image data was used to reconstruct the end-diastolic 3D geometry of the PA for all three species. The reconstruction was performed using ZIBAmira (v. 2015.28, Zuse Institute Berlin, Germany). In this study, mostly manual segmentation procedures were used. To support the manual segmentation, first, all image voxels above a specific Hounsfield Unit (HU) threshold were labelled as potential PA lumen. The threshold definition was individual due to high variability in the contrast agent concentration between the different data sets. In ovine image data, the HU threshold varied between 170 and 400. In porcine image data, the HU threshold varied between 100 and 250, whereas in human data the HU threshold varied between 80 and 190.

From these image voxel candidates, the PA lumen was reconstructed slice by slice, beginning from right ventricular outflow tract (RVOT), using brush, flood fill, as well as the region-growing tools. The segmentation was corrected by slicing through the data stack from top to bottom as well as from left to right and front to back. The voxel label field was then used to generate rough triangulated surface using a Marching Cubes algorithm. Afterward, all geometries were smoothed using a volume-preserving smoothing algorithm16 implemented in JavaView. Smoothing of the surface geometries was necessary, as the discrete resolution of CT images results in surface mesh geometries with pronounced steps. This procedure is illustrated in Fig. 1. All geometries were truncated at the MPA directly after the sinus of the pulmonary trunk, whereas all side branches of the LPA and RPA were truncated at the approximately 10 mm length from their orifices. Exemplary reconstructions for all three species are shown in Fig. 2.

Figure 1

Geometry reconstruction workflow: (A) Exemplary orthogonal slice of the CT data of a sheep. (B) Blue colour highlights the automated masking of all voxels in the slice with Hounsfield Unit values above 175. (C) Red areas indicate the PA lumen in the slice after the manual correction. (D) 3D voxel field of the PA. (E) Rough triangulated surface reconstruction. (F) Final surface after smoothing.

Figure 2
figure 2

Exemplary surfaces of the porcine (left), ovine (middle), and human (right) pulmonary arteries.

Operator-bias analysis for the surface reconstruction

Since image reconstruction was performed manually, the results are possibly affected by inter-operator variability. To quantify the impact of this potential bias, a sub-cohort of 10 patients was randomly selected for an operator bias analysis (weight: 83 ± 21.1 kg weight, age: 81 ± 4.8 years, 70% female). These data sets were reconstructed by two experienced operators who both had more than 10 years of experience in image-based reconstruction of different cardio-vascular structures. Both operators reconstructed all 10 human PA, and the resulting anatomical measurements were statistically compared.

To quantify differences between different reconstructions of the same PA, two approaches were used. First, we assessed surface differences using, by calculating the Euclidean distance for each vertex of the triangulated surface mesh of one surface to the nearest point of the second surface. During the reconstruction procedure, no alignment, transformation, or scaling is performed. Therefore, no registration of the two independent surface reconstructions was necessary. The mean and standard deviation (SD) of these distances were then calculated for a confined region including only the MPA, LPA and RPA, to avoid impact of differences in segmentation of the length of branching vessels, which were of no importance for this investigation.

Geometric parameters of interest

The in-silico platform for assessment of device effect and efficacy simulation for PAPS aims to model and predict three clinically relevant aspects: (1) interaction between blood flow and the device to assess the risk for thrombus formation; (2) device displacement, for which the patient-specific anatomy as well as the hemodynamic forces acting on the device are of interest; (3) risk for perforation of the vessel wall by the device fixation.

These aspects are all, to varying degree, affected by the PA anatomy and thus were relevant for the decision of which anatomical parameters to assess. Modelling the device fixation, an important aspect of device safety, as a dislocated device could obstruct smaller vessel segments downstream in the PA, requires accurate assessment of the diameters of the left and right PA as well as their respective curvatures. Furthermore, taper of the vessels’ segments, and the presence of side branches at the intended implantation site are of interest as well. These aspects are also important to model the forces of the device fixation onto the vessel wall. PAPS are usually manufactured from non-biological materials such as different polymers or metals. Furthermore, they form an obstacle for the blood flow in the PA, which results in a disturbed flow with increased turbulence, flow separation, and regions of flow recirculation. These disturbed flow features are known to promote thrombus formation especially in regions with low wall shear stresses (WSS)17,18.

Thus, all geometric parameters of interest for the device migration as well as the bifurcation angle between left and right PA and the enlargement of the cross-sectional area from MPA to LPA with RPA were of interest. Finally, the lengths of main left and right, as well as the main PA segments are of interest for the device implantation procedure. Table 1 summarizes motivations for the selected geometric parameters of interest in this study. Note, that the hemodynamic parameters, such as WSS or hemodynamic features, such as recirculation mentioned in the Table 1 does not represent a full list of hemodynamic parameters, which are associated with disturbed hemodynamics known to be associated with thrombus formation. A set of different hemodynamic parameters such as time-averaged WSS, oscillating shear index (OSI) or relative residence time (RRT) as well as vortical structures are parameters and features mostly associated with a prediction of the thrombosis risk by means of computational flow analysis18,19,20.

Automated measurement of geometric size parameters

The measurements of all geometric parameters were performed automatically using a centreline-based analysis of the open-ended surfaces reconstructed from the CT image data (see Fig. 2). To generate centrelines of the segmented pulmonary arteries, we used the python library vmtk22. Its algorithm approximates a given surface by fitting spheres at each point inside the surface. This way, the sample points along the centreline can be determined with their corresponding radius, which is describing the averaged distance from the sample point to the vascular wall. The amount of sample points on the centreline is resampled to a consistent number for all centrelines. Also, the indices of the bifurcation points are consistently the same, allowing automated analysis.

All edges (straight short lines connecting neighbouring vertices defining the centreline) up until the first bifurcation were considered to belong to the MPA. The LPA and RPA were defined depending on the orientation within the scanner. Starting from the first edge of the LPA and RPA, at each bifurcation, the edge with the smallest angle to the previous edge was considered to belong to the LPA/RPA respectively. All other edges were defined as side branches. For each edge, the length as well as edge-averaged diameters were obtained. The measurements were accumulated to calculate the overall length and length-weighted diameters of the MPA, LPA and RPA, respectively.

Automated measurement of geometric shape parameters

Next, parameters describing the shape complexity in a 3D space, i.e., curvature, enlargement index, taper, and the bifurcation angle, are calculated. The parameters are also described in the Table 1 and visualized in the Fig. 3. For calculation of these parameters several points of interest were specified at the centreline: the bifurcation point Pb between MPA, LPA, RPA; the end points of the LPA and RPA, and one point each on the LPA and RPA, located 10 mm downstream the bifurcation point, called P2 and P3 respectively. This downstream shift of 10 mm was chosen, as Pb was located within the vessel, while the start of the LPA and RPA should be defined using the apex of the bifurcation, i.e., when the cross sections of the LPA and RPA feature no intersections.

Figure 3
figure 3

Measurements obtained from the centreline-based analysis: the centreline (red) is subdivided into several vertices, connected by edges. For each vertex the vessel diameter is available. The vertices P2 and P4 are used for the measurement of the length L, the curvature index (CI) in the RPA using the lengths L and L0, and the RPA taper using the diameters D2 and D4. L0 is the direct distance (black line with arrows) between vertices P2 and P4, whereas L is the length of the curved centreline (red line) connecting the same vertices. The vertex Pb marks the LPA-RPA bifurcation node. The marked vertices P1, P2 and P3 are located at the distance of 10 mm from the bifurcation vertex. Diameters measured at these vertices are used to measure the enlargement index EnI. The bifurcation angle α between LPA and RPA is the angle between the vectors (black lines) connecting the Pb with P2 and P3.

The curvature index was calculated from the distance of start (P2 for the RPA and P3 for the LPA) and end node divided by the overall length of the curved vessel segment. The taper was calculated based on diameter differences at the start and end node divided by the respective segment length. The enlargement index was calculated based on the diameters of the LPA and RPA at the points P2 and P3 and the diameter of the MPA at the bifurcation point. Lastly, the bifurcation angle between LPA and RPA was determined, by calculating the angle between the vectors connecting the bifurcation point and P2 and P3 at the LPA and RPA, respectively. All individual 3D reconstructions, the centrelines as well as the geometric parameters are made available as supplemental material.

Statistical analysis

Statistical analysis was performed using IBM SPSS Statistics 28 (IBM, USA). Mean and standard deviation were reported for normally distributed parameters. Normality of distribution was assessed using a Shapiro–Wilk test. For non-normally distributed parameters, median and interquartile [IQR] range were used to report parameter distributions. A two-tailed Student’s t-Test was used to test for significant differences within normally distributed parameter differences, whereas a Mann–Whitney-U-Test and paired signed-rank Wilcoxon-Test were used for testing non-normally distributed parameter differences. To assess the agreement between both operators, Bland–Altman-analysis for the geometric parameters of interest is carried out. In addition, the intra-class correlation coefficient ICC(1,1) is calculated according to the convention by McGraw and Wong23. All tests used a standard significance level of 0.05.

Source link

Related Articles

Leave a Reply

Stay Connected

- Advertisement -spot_img

Latest Articles

%d bloggers like this: