All procedures were approved by the institutional research animal care and use committee and were compliant with regulatory guidelines (protocol number M005606). Three female Yorkshire cross domestic swine (mean weight, 54 kg) were sedated with an intramuscular administration of 7 mg/kg of tiletamine hydrochloride and zolazepam hydrochloride (Xyla-Ject, Phoenix Pharmaceutical, St. Joseph, Missouri, USA), intubated endotracheally facilitated by 0.05 mg/kg atropine (Phoenix Pharmaceutical, Burlingame, California, USA), and then underwent anesthesia induction and maintenance with 2% inhaled isoflurane (Halocarbon Laboratories, River Edge, New Jersey, USA). After being anesthetized, subjects were placed supine on the bed of a cone-beam computed tomography scanner (Artis Zee, Siemens Healthineers, Erlangen, Germany). Arterial access was obtained via a femoral arterial puncture. A vascular sheath was placed, and a 4 Fr or 5 Fr angled glide catheter was positioned in the common hepatic artery.
4D-DSA technique
As previously described [1, 2, 8,9,10,11], the 4D-DSA technique consisted of two separate C-arm rotations: the first (mask) rotation without contrast injection and the second (fill) rotation that starts prior to the contrast injection to capture the contrast inflow and time-resolved contrast kinetics. The 4D-DSA images were acquired with 6- or 12-s rotation times over an angular range of 260 degrees. Total data acquisition times (including the mask and fill rotations as well as the mid-acquisition C-arm reset) were 15 and 27 s for the 6- and 12-s rotation times, respectively. Conventional 3D reconstruction was performed to generate a constraint volume after the subtraction of mask and contrast-enhanced fill images. Time-resolved 3D image volumes corresponding to each projection image in the contrast-enhanced rotation were generated by performing constrained back-projection and normalization operations [1].
The 4D-DSA acquisitions were performed with the catheter positioned proximally in the common hepatic artery to allow adequate mixing of the contrast and blood as well as to maintain an adequate vessel segment for flow quantitation. Longer acquisition times provide more data with an increased number of cardiac cycles for image reconstruction and flow quantitation; however, they are also more prone to motion and expose the patient to a greater amount of radiation and intravenous contrast. Therefore, two frequently used acquisition times of 6 and 12 s were investigated in this study. The duration of contrast injection was dependent on the duration of the 4D-DSA acquisition. For 6-s and 12-s 4D-DSA acquisitions, 5.5-s and 11-s contrast injections were performed, respectively. The contrast injections were started 0.5-1 s after the initiation of the second gantry rotation in order to capture the inflow of contrast. For all injection protocols, iohexol 300 mgI/mL (Omnipaque 300. General Electric Healthcare, Waukesha, Wisconsin, USA) was used with a max pressure of 800 pound-force per square inch (psi). To minimize the motion-related artifacts in the reconstruction, 4D-DSA images were acquired with respiration suspended. Image reconstruction (Fig. 1) was performed using a 4D-DSA prototype provided by the manufacturer (Siemens Healthineers, Erlangen, Germany).
Assessment of optimal acquisition parameters
Quantitative flow and velocity measurements can be obtained using a previously validated 4D-DSA based algorithm described by Shaughnessy et al. [7]. This technique relies on a frequency-domain analysis of the pulsatility of the time-attenuation curves as well as 3D geometric information obtained from the iodine signal [7]. Pulsatility refers to the temporal variation in image intensity at each point along a vessel caused by temporal variations in iodine concentration. These concentration variations are a natural result of the mixing of injected contrast agent with time-varying blood flow driven by the cardiac cycle. In this study, 4D-DSA reconstructions were processed using the above algorithm, which also provides a sideband ratio (SBR) metric as a measure of contrast pulsatility in 4D-DSA [7]. The SBR is a proxy for the signal-to-noise ratio of the cardiac pulsatile waveform, and, as such, is a major determinant of flow quantification performance [7]. A higher SBR has been associated with a higher correlation seen between 4D-DSA and intravascular Doppler blood velocities [6]. Given that the objective of the study was to determine factors improving the quality of 4D-DSA reconstructions and flow quantitation, not to actually quantify hepatic arterial flow in a porcine model, SBR, not blood velocity, was the primary endpoint in the analysis.
Total of fifty-nine 4D-DSA examinations were acquired in three subjects with varying injection and acquisition protocols to understand the effect of following parameters on SBR: imaging duration (6 or 12 s), injection rate (1.0–2.5 mL/s), contrast concentration (50% or 100%), and catheter size (4 Fr or 5 Fr). Fifty-nine 4D-DSA examinations corresponded to a total of 299 data points in different arterial segments, which included various combinations of common hepatic (n = 59), gastroduodenal (n = 68), left hepatic (n = 59), left medial (n = 56), and left lateral (n = 57) arteries depending on the specific anatomy of the individual subjects or limitations with segmentation. Right hepatic arteries were not included as the right-sided porcine hepatic arterial supply is more variable and generally lacks a single dominant right hepatic artery. Given the longer course of the gastroduodenal artery without early bifurcation, for most of the acquisitions (n = 30), shorter and longer vessel segmentations were performed for gastroduodenal artery while it could not be segmented in 20 acquisitions.
The primary end point of this experiment was to assess the signal strength of contrast pulsatility (i.e., the SBR metric) with changing injection and acquisition parameters. Data points from different branches of the common hepatic artery were grouped as first-order (common hepatic), second-order (gastroduodenal or left hepatic artery), and third-order (left medial or left lateral artery) branches, and the SBR was evaluated accordingly by taking into account the inherent difference that occurs with branching distally.
Vessel cross-sectional area was selected as a secondary comparison parameter to ensure adequate contrast filling within the vessel with high-quality anatomical detail and accurate flow estimates with 4D-DSA. Therefore, for all vessels in which values for SBR were obtained, the average cross-sectional area was computed with a segmentation algorithm using an automated full width at half maximum approach at local centerline points along the vessel. A segmentation algorithm was chosen to prevent inter-rater differences that could arise with manual segmentation methods. With the remaining parameters being identical, the change in area from 1.0 to 2.5 mL/s injection rate was evaluated to assess the adequacy of filling.
The presence of reflux during injections was also noted for each acquisition to understand the influence of reflux on the SBR.
Statistical analysis
To estimate the effect of injection rate, scan time, contrast concentration, catheter size, and branch on SBR (on the natural logarithm scale), a linear mixed model was fit to the data using the ‘lme4’ package (V 1.1-1.15) in R (3.4.3) [12]. The model was adjusted for each covariate described above and allowed for interactive effects between scan time and branch number while the individual swine was modeled as a random effect to account for the influence of inter-subject differences on the SBR. Using likelihood ratio tests, it was found that two other candidate models, one which allowed for an interaction effect between injection rate and branch and one which modeled injection rate as categorical factor, were found to not fit the data significantly better than the model described above (p = 0.929, p = 0.532).
A similar linear mixed model was also fit to analyze the variance of 4D-DSA area. This model adjusted for injection rate, scan time, contrast concentration, catheter size, branch, and included an interaction between injection rate and branch. The individual swine was again modeled as a random effect.
To analyze the effect of assessed reflux on SBR, the linear mixed model described above was refit to the data but with the inclusion of a reflux term. Observations with inadvertent reflux induced via improper catheter positioning or vasospasm around the catheter tip were omitted prior to estimating the model. To assess the role of injection rate, scan time, contrast concentration, catheter size, and branch on reflux, a logistic regression model (allowing for random effects) was estimated.
After estimating each model, 95% semi-parametrically bootstrapped confidence intervals (CIs) were estimated (2000 iterations) and approximate p values were calculated using Satterthwaite’s method as implemented in the “lmerTest” package [13].
