Feasibility and accuracy of a novel automated three-dimensional ultrasonographic analysis system for abdominal aortic aneurysm: comparison with two-dimensional ultrasonography and computed tomography

We prospectively recruited 59 consecutive patients who had been diagnosed with AAA (Dmax ≥3 cm on aorta CT or 2-D US) at a single tertiary medical center between April 2016 and May 2017. The interval between aorta CT and ultrasonography was < 1 month (median, 13 days; interquartile range [IQR], 7–25 days), and all the patients underwent 2-D and 3-D US on the same day. The exclusion criteria were an estimated glomerular filtration rate of < 50 mL/min/1.73 m². The control subjects were screened from among those who underwent abdominal contrast CT for clinical reasons. Eighteen subjects who agreed to participate were included and underwent 2-D and 3-D US within 6 months after the index CT (interval: median, 96 days; IQR, 92–98 days). Overall, 59 patients with AAA and 18 control subjects were enrolled in the final analysis of 2-D US, 3-D US, and CT. This study was approved by the institutional review board of Yonsei University, Severance Hospital, Seoul, Korea. Informed consent was obtained from all the patients.

In patients with AAA, the range of imaging for AAA assessment was selected on the basis of the Dmax slice. The images of the aortic wall within 20 mm cranially and 20 mm caudally from the Dmax slice were selected for analysis (total length, 40 mm). For the control subjects, the analyzed part was selected on the basis of the left renal artery orifice. All CT and ultrasonographic variables were calculated at the site that was 10 mm lower to the left renal artery orifice, with a length of 12 mm, in the control subjects.

CT was used as the gold standard when US and CT were compared, and the entire aneurysm was displayed. The patients were scanned using a 64-section CT scanner (Sensation 64; Siemens Healthcare, Forchheim, Germany). Contrast-enhanced CT was performed using 100–140 kV with 150–220 mAs, depending on the patient’s size without electrocardiography (EGC) gating. Images were reconstructed with a slice thickness of 1.0 mm. The CT images were post-processed offline using dedicated software (Vitrea 2.0; Vital Images, Minnetonka, MN, USA; Fig. 1a). Dmax was measured perpendicular to the flow line of the vessel on 3-D reconstructed CT images [15]. The vessel, lumen, and thrombus areas were measured at the Dmax slice. The vessel volume at 40 mm was calculated with the Dmax slice located in the center of the volume. The lumen and thrombus volumes were also calculated within the same range of volume data for vessel volume measurement in the subgroup of patients with an intraluminal thrombus. A schematic representation of the measurement variables is shown in Fig. 2.

The patients did not routinely undergo any specific preparations such as fasting before the US examination. After 10 min of rest, the patients were placed in the supine position. All 2-D and 3-D US imaging acquisitions were performed using an ultrasonographic system equipped with a single-sweep volumetric transducer (RS80A, CA1-8A transducer; Samsung Medison, Seoul, Republic of Korea).

First, the 2-D US diameter and area were measured on the transverse display, from the leading edge of the adventitia anterior wall to the leading edge of the adventitia posterior wall in peak systole, on-line using a Samsung Medison ultrasonography system (RS80A; Samsung Medison). To obtain a correct anteroposterior image plane on the transverse display, the AAA was confirmed to be horizontal on the longitudinal display (Fig. 1b and c).

Then, the 3-D US acquisition was performed during breath hold (< 2 s), while the transducer was kept in a firm stable position above the cross-section showing Dmax. The image obtained was the axial volume acquisition of the AAA. For the healthy controls, 3-D US images of the field that included the renal arteries and aorta volume distal to them were obtained. Depending on the imaging results, 1–4 acquisitions were performed. The best acquisition was used for the analysis. The 3-D US acquisitions were then transferred to a workstation and later handled in experimental semi-automated 3-D software.

3-D US acquisitions were analyzed using an experimental automated software (S-3D AAA, Samsung Medison; Fig. 1d). The physicians handled each of their own 3-D US acquisitions. The software automatically detected the aortic wall to generate a centerline and 3-D aorta model. The first step was automatic delineation of the AAA wall, directly in three dimensions. The second step was automatic selection of the centerline, corresponding to the centerline of the AAA walls, not to the centerline of the AAA lumen. The third step was automatic selection of intraluminal thrombus, if present. All three steps were processed simultaneously within 2–3 s. The maximum diameter perpendicular to this centerline on 3-D US was defined as Dmax. Thus, on the orthogonal cross-section, the diameter was not restricted to a specific axis but could be established in any direction. Thrombi were automatically detected using the software. The vessel, lumen, and thrombus areas; vessel volume; and lumen volume were all measured using the same definition used in the CT analysis. Additional manual adjustments were applied for the vessel wall and thrombus-lumen interface, if needed.

Demographic characteristics are reported as percentage or mean ± standard deviation (SD). To test for differences in the observed inter-modality range of variability, and interobserver and intraobserver variabilities, the Pearson correlation coefficient was calculated for the sum and difference of the paired differences made on the same subject. Inter-modality variability was presented using Bland-Altman plots, where the differences between measurements made on the same subject were plotted against the mean outcome, showing the mean difference and the upper and lower limits of agreement given by the mean ± 1.96 × SD (2 SD). The intraclass correlation coefficient (ICC) was also calculated to evaluate the inter-modality variability, and intraobserver and interobserver reproducibility. Good correlation was defined as an ICC of > 0.8. The Student paired t test was used to compare means and mean differences.

Consensus exists that surgery or endovascular repair should be proposed when an AAA reaches a maximum diameter of 55 mm, growing rapidly over 1 cm/year in asymptomatic patients [1]. On the basis of a recent individual-based meta-analysis of trials and observational studies with repeated AAA measurements over time, intervals of 3, 2, and 1 year were proposed for AAAs of 30–39, 40–44, and 45–54 mm in diameter, respectively [9, 16].

2-D US is usually regarded as a first-line study for AAA surveillance. However, it is user dependent, and images should be skillfully acquired after angulation, especially in cases of tortuous aorta. In addition, bias exists toward smaller mean diameter measurements for 2-D US, as it measures the anteroposterior diameter with lower inter-measurement variability [10]. By contrast, measurement of the aorta with 3-D US does not require the assumption of any specific geometry but can be directly computed by manual or automatic segmentation of continuous slice data, which can overcome the limitations of 2-D US and would be the optimal method for AAA surveillance. Therefore, several studies have developed 3-D US-based analyses for AAA [17, 18]; however, until now, no standardized methods for acquisition and analysis have been established.

Here, we propose a novel approach to extract more information from 3-D US acquisitions obtained with a single-sweep volumetric probe by combining the recorded volume with a dedicated post-processing software that makes it possible to extract the surface of the AAA and its centerline. The novelty of this study is that it shows the possibility of automatically extracting the Dmax of the AAA in any direction, perpendicular to the centerline, from the 3-D US AAA segmentation within several seconds. The clinical interest could be to provide a unified definition of Dmax in any direction using 3-D US. Improving the quality of measurements is aimed primarily at improving patient care over the successive stages of the disease, including screening, decision for intervention, and follow-up. Using this 3-D US software, we correctly categorized 96.6% of the patients with the same CT classification for the recommended follow-up intervals, whereas the 2-D US classification matched only in 59.2% of the patients. Therefore, our 3-D analysis system can be used as a first-line modality for AAA screening and surveillance, as it is inexpensive and noninvasive and can screen quickly and easily.

As AAA is a 3-D disease, the clinical interest in AAA volume measurement lies in enabling better prediction of the evolution of small AAAs and of AAAs post endovascular AAA repair. AAA volume estimation has previously been reported with CT acquisitions combined with post-processing [19‐22]. Nevertheless, this technique is not routinely used in clinical practice because, compared with US, CT presents drawbacks such as exposure to radiation, injection of iodine contrast medium, and higher costs.

Our novel method automatically calculated the aorta volume. In addition, the novelty of this software is that it automatically quantifies thrombus volume using dedicated software, in addition to the whole vessel volume and area. The impact of intraluminal thrombus on AAA rupture risk is controversial until now. Although several studies have demonstrated that intraluminal thrombus reduces aortic wall stress, thereby acting as a mechanical buffer, [23, 24] increasing evidence shows that high intraluminal thrombus burden may be a surrogate marker of decreased aortic wall strength and a characteristic of high-risk small aneurysms [25, 26]. Our novel software can provide additional information on intraluminal thrombus volume by automatic measurement and therefore facilitate the unraveling of the role of intraluminal thrombus volume that cannot be acquired with 2-D US.

The main limitation of the present study was that because of the probe characteristics, information about the neck was not available for large AAAs. Further development of the hardware and scanning method is expected to facilitate data acquisition and enhance the performance of the system. US imaging of the abdominal aorta may be technically limited because of patient obesity or excessive abdominal gas, even though we did not encounter any technical failure of 3-D US in our study. However, with appropriate patient preparations such as oral intake restriction for at least 8 h before the examination, many of the body habitus pitfalls may be overcome. The mean body mass index of the patients was significantly lower than those of other populations such as US citizens. As obesity decreases the quality of US imaging, our results may not be applicable to other populations with a higher prevalence of obesity. Both CT and US images were acquired without ECG gating in the present study, and changes in AAA according to cardiac cycle were not considered. However, the changes in the abdominal aorta during the cardiac cycle are less significant than those in the thoracic aorta, and abdominal aorta CT is usually performed without ECG gating in routine clinical practice. Our intention was to investigate the accuracy of 3-D US compared with that of CT, and we applied the same protocol to CT and 3-D US without ECG gating. Further studies are warranted to evaluate the clinical value of the cyclic change evaluation.