Fifty randomly selected patients from the Leiden University Medical Center, the Netherlands, who had chest pain complaints and were referred for a CCTA were included in the current study. Two different phase reconstructions from the same scan from each patient were chosen; the two reconstructions were in the range of either 70–80% or 30–80% for the entire cohort. In principle, this meant that plaque thickness differences should have been absent, as both phases were made almost simultaneously. The compared reconstructed phases were always within the same RR interval, which constitutes the time between two successive R waves of the QRS signal on the electrocardiogram (ECG). The compared phase pairs were always within the same gated window; either 70–80% or 30–80%, and always constituted a 75% phase and a randomly reconstructed other phase. All data were clinically acquired and retrospectively analyzed. The institutional review board of the Leiden University Medical Center, the Netherlands, approved this retrospective evaluation of clinically collected data and waived the need for written informed consent. This study was performed in accordance with the Helsinki Declaration of 1964 and its later amendments.

CCTA was performed using a 320-row volumetric scanner (Aquilion ONE and Aquilion ONE Genesis Edition, Canon Medical Systems, Otawara, Japan). Heart rate and blood pressure were monitored 1 h before CCTA. Metoprolol (from 25 mg up to 150 mg) was administered orally to patients exceeding a heart rate of 60 beats per minute (bpm) provided that no contraindications were present. Additional metoprolol was injected intravenously if the heart rate remained above 60 bpm. Nitroglycerin (0.4 mg) was administered sublingually 4 min prior to CCTA. The scan parameters were as follows: a detector collimation of 320 × 0.5 mm, a 275-ms gantry rotation time, and a temporal resolution of 137 ms for the Aquilion ONE Genesis Edition; a detector collimation of 320 × 0.5 mm, a 350-ms gantry rotation time, and a temporal resolution of 175 ms for the Aquilion ONE. The peak tube voltage was 100–135 kV with a tube current of 140–580 mA for both scanners. 70–80% of the RR interval was scanned using prospective ECG triggering. When the heart rate was above 65 bpm, 30–80% of the RR interval was scanned. The first 50–90 ml of contrast agent (Iomeron 400, Bracco, Milan, Italy) was administered in the antecubital vein. Thereafter, 20 ml of a 1:1 mixture of contrast and saline and finally 25 ml of saline were administered. CCTA was performed at the next beat when the threshold of 300 Hounsfield units (HU) was reached in the descending aorta. The protocol settings were the same for the Aquilion ONE and Aquilion ONE Genesis Edition; a tube voltage of 100 kV was generally used. A 120-kV tube voltage was used for patients who had a weight exceeding 130 kg and/or were bearing an implantable cardioverter-defibrillator (ICD). Tube current ranged between 300 and 900 mA depending on patient size. Field of view (FOV) was also dependent on patient size and ranged between 200 and 280 mm. Image reconstruction was done using iterative reconstruction by means of adaptive iterative dose reduction-3D (AIDR-3D) enhanced for the Aquilion ONE Genesis Edition and AIDR-3D for the Aquilion ONE using the FC03 reconstruction kernel for both scanners. Iterative reconstruction strength was set at mild, standard, or strong depending on the image noise. Image size was set at 512 × 512. The slice thickness of the reconstruction was 0.25 mm for all but two of the reconstructed phases, which had a slice thickness of 1.0 mm.

It is important to note that the protocol and image reconstruction settings remained consistent for all compared reconstructed phases.

Dicom images were transferred to an offline workstation for analysis. Dedicated software (QAngio CT Research Edition v3.1.5.1, Medis Medical Imaging, Leiden, the Netherlands) was employed to conduct automatic tracing of the coronary arteries and the semi-automatic detection of the lumen and vessel wall contours. The contours were corrected manually if needed, whilst the reader was blinded to the results of the other phase. Coronary artery tree extraction and vessel selection are depicted in Fig. 2.

The complete coronary tree is extracted from the CCTA. In this example, the left anterior descending artery (LAD) is marked in blue for performing plaque delineation. LM left main artery, pLAD proximal left anterior descending artery, dLAD distal left anterior descending artery, pRCA proximal right coronary artery, pCX proximal circumflex artery, LCX left circumflex artery, D1 first diagonal artery, OM1 first obtuse marginal artery, mLAD mid left anterior descending artery, CCTA coronary computed tomography angiography

A software program developed in house by Cao et al. [7] was employed to extract the three-dimensional (3D) lumen and vessel wall surface models of the three main arteries in each of the two scans. The software co-registers both 3D models and encodes the local plaque thickness differences between the two scans on the surface of a model. Subsequently, ParaView (version 5.9.0) was utilized for the 3D visualization of the generated models.

In order to quantify image quality, the CNR was calculated separately for the left anterior descending artery (LAD), the right coronary artery (RCA), and the circumflex artery (Cx). We opted to use CNR as a metric to quantify image quality as this has been proven to affect the accuracy of CCTA. Furthermore, it has been demonstrated that a reduced CNR results in a reduced sharpness of vessel visualization. The latter negatively influences plaque visualization and thus also software-aided plaque delineation [11, 12]. Contrary to the signal-to-noise ratio (SNR), CNR serves as a quantitative metric for low-contrast lesion detection: the higher the CNR between lesion and background, the more likely the detection of the lesion [13]. Although the SNR and CNR formulas are similar, SNR lacks specificity, as it does not consider the mean intensity of the surrounding epicardial tissue [14]. Therefore, CNR presents superior significance in contrast-enhanced scans like CCTA, as it is a measure of image quality based on a contrast [15]. A total of seven regions of interest (ROIs) per patient were defined for the measurement of the intensity values and the subsequent calculation of the CNR. The first ROI was placed in the ascending aorta, superior and in close proximity to the origin of the RCA, to define image noise. Thereafter, three ROIs were placed in the most proximal part of each coronary vessel. The final three ROIs were placed in the epicardial tissue surrounding each vessel, adhering to the same slice position and in spatial proximity to the ROI in the corresponding vessel. ROI placement was performed meticulously to exclude calcifications, plaques, vessel walls, and any potential image artifacts. Figure 3 depicts an example of a patient with ROIs placed in the aorta, LAD, and surrounding epicardial tissue.

Regions of interest are manually drawn in the aorta (A), proximal LAD (B), and the corresponding epicardial tissue surrounding the LAD (C). This means of operation is the same for the Cx and the RCA. LAD left anterior descending artery, Cx circumflex artery RCA right coronary artery

The CNR was subsequently calculated for each vessel using the following formula:

$$\mathrm=\frac_}- _}}_}},$$

in which \(_}\) represents the mean HU intensity of the specific coronary vessel, \(_}\) represents the mean HU intensity of the epicardial tissue in spatial proximity to the specific coronary vessel, and \(_}\) represents the standard deviation of the HU intensity in the ascending aorta.

Coronary lumen and vessel wall contours are detected in the multi‐planar reformatted images of the artery. Based on the detected lumen and vessel wall contours, the plaque thickness at a certain location in an artery can be calculated. This is done by calculating the distance between the points at which the lumen contour and the vessel contour insersect with the line through the lumen center. The change in plaque thickness is determined as the difference in plaque thickness at the corresponding location between scans [7]. It is important to note that the accuracy of contours and thus plaque delineation is dependent on the scan quality [16]. Therefore, thresholds are needed to filter out insignificant changes in plaque thickness differences resulting from variations in contour quality. Figure 4 depicts a clinical example of a case with plaque progression in the LAD that shows the importance of using thresholds for plaque thickness change visualization.

A newly formed plaque is observed in the proximal LAD, as marked by the blue arrow (A). No other vessels have plaque (changes). Multiple areas are identified as having plaque progression using cutoff values of − 0.5 and 0.5 (B). Larger cutoff values of − 0.75 and 0.75 still do not allow plaque progression to be discerned in the RCA and the middle part of the LAD, as marked by the red areas (C). Finally, cutoff values of − 1.0 and 1.0 seem to correlate well with the visual observations in panel A (D). This demonstrates the importance of using cutoff values, yet the adaptive values must still be calculated using the CNR as a marker of scan quality. Plaque thickness differences are given in mm. BA baseline, FU follow-up, RCA right coronary artery, LAD left anterior descending artery, Cx circumflex artery, CNR contrast-to-noise ratio

In order to establish vessel-specific thresholds, calibration graphs were created between the lowest measured CNR of a vessel in both phases and the largest negative and largest positive differences in plaque thickness measurements between two-phase scans. For each patient, two different reconstructed phases from the same scan were compared. As plaque differences between two reconstructed phases from the same scan and from the same patient should always be zero, it is possible to compare both phases in a two-way manner. Hence, for each patient, two values of the plaque thickness difference were obtained, yielding a total of 100 values. Subsequently, any plaque thickness delineation differences between two-phase scan sets had to be attributable to different factors such as scan quality. The software tool from Cao et al. [7] was utilized for automatically calculating the negative and positive plaque thickness differences. Subsequently, the largest negative and largest positive thickness differences were plotted against the vessel-specific CNR. Linear regression facilitates the determination of the linear relationship between a dependent and independent variable, in this case plaque thickness difference and CNR, respectively. Formulas were derived through linear regression analysis conducted on the aforementioned charts using SPSS software (version 25, SPSS IBM Corp., Armonk, New York). The standard error of the estimate which is used in linear regression analysis was multiplied by a value of one instead of the customary two. This was done pragmatically in order to ensure that the model was capable of detecting relatively small plaque changes with regard to the average coronary lumen diameter, which is between 3 and 4 mm [17]. A detailed step-by-step flowchart depicting the aforementioned process is presented in Fig. 5.

Flowchart depicting the process of creating formulas for thresholds of plaque differences using scan quality. ROI region of interest, CNR contrast-to-noise ratio, RCA right coronary artery, LAD left anterior descending artery, Cx circumflex artery

A random set of 15 scans were utilized for inter-observer measurements, resulting in the analysis of 45 coronary vessels. Observer AB (with 13 years of experience in cardiovascular image analysis) also drew a total of seven ROIs per patient for CNR measurements. Thereafter, the calculated CNR values were compared to those obtained by observer FY (with 3 years of experience in cardiovascular image analysis). Subsequently, correlations were tested using Pearson’s correlation coefficient.