Phantom
The phantom was composed of a radiopaque skeleton and a 3D vascular model (Fig. 1). The vascular model was created from a CTA dataset of a patient with an infrarenal abdominal aortic aneurysm (female, 174 cm, 72 kg, 82 years). The patient agreed to the use of her data via written informed consent. The phantom was produced by hybrid additive manufacturing based on fused deposition modeling (Felix 3, FELIXprinters), whereby the inner contours of the vessels were printed with a water-soluble material and then the model was covered with silicone (Shore A 37). The vessels represented include the aorta, supra-aortic, visceral, and iliac vessels. Vascular access was possible via the ascending aorta, the supra-aortic, and both external iliac vessels. The Phantom was connected to a diaphragm dosing pump, simulating pulsatile flow (Sigma, ProMinent®DeutschlandGmbH, Heidelberg). Breathing motion was not simulated.
Computed tomography angiography
The CTA was performed on a 128-slice CT scanner (Somatom Definition AS+®, Siemens Healthcare, Erlangen). Imaging parameters were as follows: tube voltage = 120 kV; reference tube current-time product = 200mAs; rotation time = 0.3 s; collimation = 0.6 mm. The phantom was placed in the supine position and filled with an iodine CM (mixing ratio CM/0.9% sodium chloride 1/15; Imeron 300®, Bracco). Images were reconstructed employing a soft tissue kernel (B30f) and an effective slice thickness of 1.0 mm.
Magnetic resonance angiography
A standard cartesian 3D fast field echo MR angiography was performed using a 3 T scanner (Philips Ingenia Omega dStream, Philips, Best) with a 20-channel body surface coil. Data were acquired in the axial plane for the thoracic and the abdominal regions. Acquisition parameters were as follows: field of view = 380x462x200mm; slice thickness = 5 mm; image matrix = 292 × 330; and time to repeat/time to echo = 3.8/2.4 ms. An acquired voxel size of 1.3 × 1.4 × 2.4 mm was reconstructed to 0.6 × 0.6 × 1.2 mm. The phantom contained blood mimicking fluid, consisting of 36.6% glycerine in 0.9% sodium chloride solution doped with CM (Gadubotrol, Gadovist®, Bayer, Leverkusen).
Fluoroscopy and fusion image processing
Images were imported to a workstation in the angiography suite (Allura Xpert® FD20/15, 3.4, Phillips, Best). The following two steps had to be performed with dedicated software (VesselNavigator®, Phillips).
Planning
Vessel segmentation from CTA/MRA data was conducted semi-automatically and corrected manually, if necessary. The orifices of the vessel ramifications were marked via circular landmarks. Ideal placement of such a marker is shown in Fig. 2.
Registration
2D-3D fusion was performed by employing two fluoroscopic images in 90° right anterior oblique (RAO) and anterior-posterior projection. The 3D vessel model was consecutively fitted to fluoroscopic images using osseous landmarks.
Before 3D-3D fusion, a cone-beam CT was acquired in the angio-suite. Image fusion was achieved by correlating vessel-specific landmarks in both CT datasets (cone-beam CT; pre-interventional CT) as follows: In the thorax, ramifications of the supra-aortic vessels were used. In the abdomen, orifices of the visceral arteries were employed for alignment.
To test for inter-reader reproducibility, image fusion with each technique and image modality was performed by two investigators, blinded to the results of the other investigator. For the intra-reader analysis, the measurements were repeated by one investigator at an interval of 6 weeks.
Data analysis
To evaluate fusion quality, fluoroscopy of a full c-arm rotation from RAO to left anterior oblique (LAO; 90° to − 90°) was recorded at two levels of the phantom (thorax and abdomen). A custom MATLAB tool was used, which was validated using centerlines defined manually by three experts, based on the methodology presented here (Schaap et al. 2009). The mean error of the tool for centerline definition was 0.56 ± 0.34 mm. The tool calculated two centerlines after manually outlining the actual vessel’s borders and the virtual 3D vessel model on fluoroscopy images (Fig. 3). Deviation of these two centerlines was measured on a pixel-wise basis every 10 mm for seven c-arm angulations (− 90°, − 60°, − 30°, 0°, 30°, 60°, and 90°). The centerline defined on the fluoroscopy images served as reference standard for all accuracy measurements.
Accuracy of the landmark placement was evaluated on a binominal basis, as shown previously (Schwein et al. 2018; Chinnadurai et al. 2016). Landmarks were placed at each ramification of the aorta. A soft guidewire (GLIDEWIRE®, Terumo) was placed in the adjacent vessel, and a score of 0–1 was given, depending on the position of the wire (either in- or outside the circular marker, Fig. 2e).
Statistics
Statistical analyses were performed using SPSS® (version 25.0, IBM Corp.).
Deviations of centerlines are presented as mean ± standard deviation. Differences between registration techniques and imaging modalities were tested for significance using student’s t-test. Significance was accepted at a p-value of < 0.05.
Graphs were calculated to illustrate the deviation of the centerlines for all c-arm angulations over the vessel’s course against the reference standard. Bland–Altman analysis, including calculation of mean bias and limits of agreement (mean bias±1.96*standard deviation), was performed to assess the differences between imaging modalities in head-to-head comparison.
To test for inter- and intra-reader reproducibility, the intraclass correlation coefficient (ICC) with 95% confidence intervals was calculated.