Datasets of 12 patients with an ECAA all located in the internal carotid artery (ICA) were retrieved from our Carotid Aneurysm Registry (www.carotidaneurysmregistry.com) . The registry has been approved by the local ethics committee, and all patients gave informed consent. For the purpose of this study, a computed tomography angiography (CTA) scan with slice thickness below 1.0 mm was eligible for inclusion to guarantee proper slice thickness for reconstruction. This necessary condition limited the amount of eligible CTAs to 12 due to rarity of disease. The CTAs had been performed for evaluation or treatment of ECAA between 2008 and 2017, in the University Medical Center Utrecht. We aimed to select an equal amount of cases with fusiform and saccular ECAA. As specified within the registry protocol , fusiform aneurysms were defined as ≥ 150% diameter increase of the normal ICA diameter, while saccular aneurysms were defined as a distended sac of any size affecting only part of the ICA circumference.
A 64-slice or 128-slice CT scanner (Philips Brilliance; Philips medical systems, Best, the Netherlands) was used to acquire the CTA scans. The carotid arteries were visualized from the aortic arch to the skull base. Median slice thickness was 0.67 mm (range 0.62–0.90 mm), increment 0.33, collimation 64 × 0.625, and pitch 0.609. Radiation exposure parameters were 100–120 kV and 150–300 mA s. The field of view is set per patient. Injection of 65 ml intravascular contrast (ultravist 300, Schering, Berlin, Germany) was followed by a saline bolus of 40 ml, both at a flow rate of 6 ml per second.
Our study focused on the evaluation of semiautomatic measurement software packages. A search was performed to identify software packages which facilitated semiautomatic vessel tortuosity measurements and were commercially available. To this end, the MEDLINE database was searched using the search terms “software,” “length” or “tortuosity,” “vascular,” and synonyms. Availability of free trial licenses was required in order to participate in this comparative study. Four commonly used commercial software packages were selected: 3mensio Vascular (version 8.1, Pie Medical Imaging BV, Maastricht, the Netherlands), Aquarius iNtuition (version 126.96.36.1995, iNtuition Cloud, TeraRecon, Foster City, CA, USA), Vitrea (version 7.4, Vital Images Inc., Toshiba Medical, Minnetonka, MN, USA), and Aycan OsiriX PRO (version 3.10.xxx, Aycan Medical Systems, Rochester, NY, USA). All software packages are commonly used for semiautomatic (vessel) analysis and centerline composition [16,17,18,19,20,21,22,23,24,25].
Two observers (EEV and VECP) independently scored the 12 datasets at two time points (round 1 and 2, interval ≥ 1 week) with the four software packages. Observers were blinded to each other’s measurements and to earlier measurements with the same software package. For each software package, both observers received a training session by the company of 1 h, and practiced three measurements in order to familiarize with the package and overcome the early learning curve.
In all software programs, carotid artery tortuosity was determined by calculating the tortuosity index (TI) of the carotid artery ipsilateral to the ECAA. The TI was defined as the true length of the central luminal line (CLL) divided by the straight distance. It was calculated in two ways: from the skull base (just proximal from the carotid siphon) until (1) the carotid bifurcation and (2) aortic arch (Fig. 1).
The primary outcome measure was the reproducibility of tortuosity measurements, expressed as the inter- and intra-observer variability in the TI, as measured with the different software programs. The secondary outcome measure was defined as the agreement in absolute TI between the software programs. For both the primary and secondary outcome measures, the correlation between measurements was calculated by using the TI from the skull base to the carotid bifurcation, since ECAAs are located primarily in the ICA. The tertiary outcome measure was the time needed per scan (difference in time between scoring round 1 and 2 (learning curve)).
Inter- and intra-observer variability
The intraclass correlation coefficient (ICC) was used to calculate the inter- and intra-observer variability for measurements obtained with one software package (model: two-way mixed, type: consistency). An ICC of 1.0 equals perfect agreement, an ICC of 0.81–0.99 excellent agreement, and an ICC of 0.61–0.80 substantial agreement . The first round of measurements of both investigators was compared in order to calculate inter-observer variability for each software package. Bland-Altman plots were constructed to assess presence of systematic differences between both observers.
Agreement on absolute TI
The ICC was also used to calculate agreements on obtained TIs per software package. In order to calculate the differences in measured absolute TI with each software package, the average TI (TIaverage) per case was calculated by taking the average of all four measurements. This was done for each software package separately, thereby producing one TIaverage per software package for each of the 12 cases.
Time needed per scan and learning curve
A Kruskal-Wallis test was used to calculate differences between software packages in time needed to complete all measurements, while a Wilcoxon signed rank test was used to calculate differences in time needed to complete round 1 and 2 within one software package. Mann-Whitney U tests were used as post-hoc tests, and Bonferroni correction was applied to account for multiple testing.
Statistical analyses were conducted using SPSS 22.0 (IBM Corp. Released 2013. IBM SPSS Statistics for Windows, Version 22.0. Armonk, NY: IBM Corp.). A P value less than 0.05 was considered statistically significant.