CT analysis

3.8.1 Features associated with the napkin-ring sign

As the most advanced, model-based image reconstruction technique is available mostly in the research phase, the dataset used for plaque feature assessment was reconstructed using ASIR technique. The purpose of this sub-study was to define features of advanced atherosclerotic plaques that promote or interfere with the delineation of the NRS in coronary CT angiography, thus, only cross-sections containing advanced plaques as classified by histology were used. The corresponding CT images were reviewed in consensus by two radiologists with 6 and 10 years of experience in coronary CT angiography. All reading was performed using a fixed window setting [700 Hounsfield Units (HU) width, 200 HU level]. Based on the CT appearance, it was specified whether a napkin-ring sign could be identified within the plaque. The NRS was defined by a low-attenuation plaque core surrounded by a circumferential area of higher attenuation (45). Additionally, we measured the attenuation of the non-calcified plaque portion of all plaques using an in-house program developed in Matlab. For this purpose, the outer circumference of the vessel and the lumen were manually traced and the median density of the pixels within the plaque was calculated. In plaques with a positive NRS, we also measured the median density of the central hypodense area and the peripheral hyperdense ring.

3.8.2 Image quality of different reconstruction techniques

Image analysis was performed at a commercially available workstation (Advantage Windows 4.2; GE Healthcare, Milwaukee, WI, USA) by two experienced observer blinded to the image review results. The three image sets, obtained with reconstruction techniques FBPR, ASIR, and MBIR in each patient, were displayed side by side with a preset soft tissue window (window width, 200 HU; window level, 700 HU).

3.8.2.1 Qualitative image analysis

The two reviewers were asked to grade the overall image quality on a four-point scale:

1 = excellent; 2 = good (minor artifacts); 3 = moderate (considerable artifacts but diagnostic quality); 4 = poor, non-diagnostic. Regarding calcification and blooming artifacts, both readers were asked to rate if there is calcification or not. If present, a three-point scale was used to qualify blooming artifacts: 0 = no blooming; 1 = blooming but lumen clearly visualized; 2 = blooming, no lumen visible. Image noise was defined as overall graininess or mottle in the coronary artery. It was graded on a four-point scale: 1 = no image noise; 2 = average noise; 3 = above average noise; 4 = severe noise.

Image sharpness was evaluated on a five-point scale: 0 = extremely poor, no definable lumen; 1 = poor, severe streaking comprising more than 50% of the vessel lumen;

2 = fair, vessel walls may show mild streaking or blurring, but a defined lumen is identifiable; 3 = good, coronary walls may be blurred but lumen is clearly discernible and separate from the surrounding fat; 4 = excellent, coronary walls are sharp and well-defined with no blurring. Such an assessment scheme was used in a previous study of coronary lumen analysis at CT (76,77).

3.8.2.2 Quantitative image analysis

We obtained mean CT attenuation values (in Hounsfield units) for the peri-coronary fat and coronary artery by manually placing circular regions of interest (ROIs) at the same image level. The attenuation of the coronary artery was recorded from a single drawn ROI that was as large as the vessel lumen. For each protocol, image noise was measured as the standard deviation of the pixel values from a circular or ovoid ROI drawn in a homogeneous region of the peri-coronary fat. For all measurements, the size, shape, and position of the ROIs were kept constant among the three protocols by applying a copy and paste function at the workstation. The attenuation of the peri-coronary fat was recorded as the measurement of one ROI placed in the peri-coronary area to the coronary lumen. Areas of focal changes in parenchymal attenuation and prominent artifacts, if any, were carefully avoided. For each of the three reconstructions, the

contrast-to-noise ratio (CNR) relative to peri-coronary fat for the coronary artery was calculated by using the following equation: CNR = (HUlumen – HUfat) / SDlumen, where HUlumen and HUfat are the mean CT numbers of the coronary artery lumen and the peri-coronary adipose tissue, respectively. SDlumen represents the standard deviation of luminal CT number.

3.8.3 Automated plaque assessment

For each vessel reconstructed by three different algorithms (FBPR, ASIR, MBIR), a separated 3D dataset was generated. All images were anonymized regarding the applied reconstruction algorithm and vessel origin. All 3D datasets were transferred to a dedicated offline workstation (Vitrea Advanced Cardiac Solutions, Vital Images, Minnetonka, MN, USA), which allowed automated vessel segmentation and quantification of coronary artery components (49,78).

The starting point for the luminal centerline was manually set at the proximal ending of the vessel by a coronary CTA reader in the 3D dataset and visually verified. Afterwards, the software performed the automatic vessel segmentation and fitted the inner and outer vessel-wall boundaries. The automatically fitted boundaries were reviewed by an experienced coronary CTA reader on cross-sectional images with 0.5 mm increments. If the boundaries did not follow the anatomical structures, the inner and/or outer vessel-wall boundary was manually corrected. The proximal 40 mm of each vessel (left anterior descending [LAD], left circumflex [LCX], right coronary [RCA] artery) starting from the plastic luer was included in the assessment, plus the left main (LM) which was counted as part of the LAD. All measurements were exported for further processing.

The primary endpoint was the percentage of cross-sections where a manual correction of the automatically fitted vessel-wall boundaries (inner or outer) was necessary to perform. Secondary endpoints were the following: 1.) Percentage of cross-sections that required corrections of only the inner vessel-wall boundary; 2.) Percentage of cross-sections that required corrections of only the outer vessel-wall boundary; 3.) Benefit of MBIR over FBPR (or over ASIR), which was defined as cross-sections that needed any

corrections on FBPR (or ASIR) but not in MBIR, as compared to cross-sections, which showed no difference between MBIR and FBPR (or ASIR). The latter group contained cross-sections which have been or which have not been corrected in both image reconstruction algorithms (MBIR and FBPR; MBIR and ASIR).

The reproducibility of the primary endpoint was confirmed in a random subset (33% of the entire cohort). The vessel assessment including the boundary delineation was performed twice with a time gap of two weeks. A high correlation was observed with respect to the percentage of corrected cross-sections per vessel between the two assessments (intraclass correlation coefficient: 0.99). Also, an excellent regional agreement was achieved comparing which individual cross-sections were corrected the first and second time (kappa: 0.86).