This research protocol was approved by the Institutional Review Board, and informed consent was obtained from all patients before examination in this prospective study. This study enrolled patients with suspected coronary artery diseases who had undergone CCTA from August 2016 to January 2017, in our institution. Data were collected from subjects having single-vessel disease with non-calcified coronary plaques and significant stenosis (≥ 50% in diameter) on CCTA. Exclusion criteria for CCTA included subjects with renal dysfunction (glomerular filtration rate < 60 mL/min), arrhythmia, known allergy to iodinated contrast media, and the inability to sustain a 10-second breath-hold. Finally, 66 subjects (53 men and 13 women; aged 39–76 years; mean age, 55 ± 13 years) were included.

CCTA was performed on a 256-slice CT scanner (Brilliance iCT; Philips Healthcare, Cleveland, OH, USA). In all patients with a heart rate (HR) > 65 beats per minute (bpm), a β-blocker (metoprolol tartrate; AstraZeneca, London, UK) was orally administered at the dose of 25–50 mg to achieve the target HR ≤ 65 bpm, 2 hours before CT examinations. Administration of sublingual nitroglycerin (LGC, Teddington, UK) for coronary vasodilation was performed at the time of imaging. CCTA was performed using a prospective electrocardiogram (ECG) triggering protocol using the following parameters: slice configuration, 128 × 0.625 mm; gantry rotation time, 270 ms; pitch, 0.18; tube voltage, 100 kVp; effective tube current time product, 600 mAs with ECG-dependent tube current modulation; and slice thickness, 0.8 mm with slice increment of 0.45 mm. A bolus-tracking technique was used to monitor the CT value of the ascending aorta with a trigger threshold of 120 HU. The scans were started 5 seconds after the trigger. Each patient received 70 mL of contrast media (iopromide, Ultravist® 370; Bayer Healthcare, Berlin, Germany) intravenously at the speed of 5 mL/s, followed by 40 mL of saline at the same injection speed.

The CT data were respectively reconstructed with an HIR algorithm (iDose4-level; Philips Healthcare) and an implementation of the new knowledge-based IR algorithm (IMR-level 1) for each subject. The reconstructed parameters were the same, including 0.8 mm thickness with 0.45 mm increment, 512 × 512 in matrix, 250 mm field of view and a standard cardiac reconstruction kernel (XCB). All CCTA data were transferred to a computer workstation (Extended Brilliance WorkspaceV4.5.2; Philips Healthcare).

Maximum intensity projection (MIP), multi-planar reconstruction (MPR), and curved-planar reconstruction images were reconstructed and used for detecting coronary plaque by a radiologist with 5 years of experience in CCTA. An MIP image of the long axis of the coronary artery was reconstructed to identify lesions with significant stenosis, and then cross-sectional MPR images of 1 mm thickness through the plaque were reconstructed for demonstrating the plaque and measuring the contrast-to-noise ratio (CNR) between the plaque and adjacent adipose tissue. The CT value of the aortic root was measured on axial source image with both HIR and IMR algorithms. The region of interest (ROI) in the aortic root was about 200 mm². The CT values were also measured in the lumens of the left main (LM) trunk, proximal, middle, and distal segments of the right coronary artery (RCA), left anterior descending (LAD) artery, and proximal and distal segments of the left circumflex (LCX) artery. The round ROI was placed in the center of the vessels and was neither small enough to be affected by pixel variability nor large enough to approach the edge of the vessels (Figs. 1, 2). The CT values of plaques were measured on both HIR and IMR-reformatted axial images using ROIs with the same size and location. Each ROI was located at the lowest attenuation area of the plaque and drawn as large as possible. Microcalcifications or other disturbing structures were carefully avoided. The CT values of the adjacent perivascular adipose tissues were also measured (Fig. 3). Noise was defined as the standard deviation (SD) measured in the background (air), anterior to the patients' chest wall. The area of the ROI for measuring noise was about 50 mm² (Fig. 4). The measurement in each ROI was performed three times, and the average value was calculated. The signal-to-noise ratio (SNR) was calculated for intravascular attenuation and plaque attenuation. CNR was calculated as follows: ([attenuation of vessel lumen or attenuation of plaque] − [attenuation of the adjacent perivascular adipose tissue]) / image noise. The coronary segmentation was determined according to the Society of Cardiovascular CT guidelines for the interpretation and reporting of CCTA (16).

High-risk plaque analysis was performed using the follow criteria: low-attenuation plaque (plaque attenuation lower than 60 HU), napkin-ring sign, spotty calcifications (calcification < 3 mm in size), and positive remodeling. Napkin-ring sign was defined as an outer high-attenuation rim (< 200 HU) with an inner hypo-attenuation area (< 130 HU), which was not adjacent to a calcification and present on a minimum of two adjacent 1 mm axial slices. We carefully checked the non-contrast CT for microcalcifications when we evaluated a napkin-ring sign. The remodeling index was calculated as the ratio of the maximal cross-sectional vessel diameter, including the plaque and lumen, and its closest proximal normal reference vessel diameter (17). A remodeling index > 1.05 was considered as positive remodeling (18).

Two experienced radiologists with 10 and 5 years of experience with CCTA, respectively, blinded to image algorithms, identified plaque characteristics independently and collectively resolved any differences in their readings.

Continuous variables were presented as mean ± SD, and categorical variables were expressed as frequencies or percentages. A paired t test was used to determine the differences in image noise, vascular attenuation, plaque attenuation, adipose tissue attenuation surrounding plaque, SNR, and CNR between the two image reconstruction methods. Chi-square test was performed to compare the difference between the HIR and IMR algorithms in detecting plaque characteristics. Cohen's kappa test was used to analyze the inter-observer agreement in discerning the plaque characteristics. The κ values were defined as follows: 0–0.2, poor agreement; 0.21–0.40, fair agreement; 0.41–0.60, moderate agreement; 0.61–0.80, substantial agreement; and 0.81–1.00, almost perfect agreement. A two-tailed p value < 0.05 was considered statistically significant. All statistical analyses were conducted by SPSS 18.0 for Windows (SPSS Inc., Chicago, IL, USA).

Seventy-seven non-calcified or predominantly non-calcified plaques with significant coronary stenosis were detected. Two plaques were located in LM, 51 in LAD, 23 in RCA, and 1 in LCX.

A total of 72 (93.5%) high-risk plaques were detected on IMR images, including 42 (54.5%) showing low attenuation, 40 (51.9%) with napkin-ring sign (Figs. 5, 6), 18 (23.4%) with spotty calcifications, and 59 (76.6%) with positive remodeling. There were 70 (90.9%) high-risk plaques identified on HIR images, including 41 (53.2%) showing low attenuation, 19 (24.7%) with napkin-ring sign, 18 (23.4%) with spotty calcifications, and 57 (74.0%) with positive remodeling. The chi-square test showed no statistically significant difference in the presentation of low attenuation plaque, spotty calcification, or positive remodeling between the HIR and IMR reconstruction methods, but there was a statistically significant difference in the discernment of the napkin-ring sign between the two algorithms (χ² = 12.12, p < 0.001) (Table 1).

Table 2 summarizes the results of image quality between the HIR and IMR reconstruction algorithms. The noise with IMR was significantly lower than that with HIR (p < 0.001). The attenuation of the aortic root with IMR was not different from that with the HIR reconstruction method (p = 0.674). The CT values of the coronary artery segments measured using IMR were significantly greater compared to those using HIR (p < 0.001). The CT values of adipose tissues surrounding the vessels measured using IMR were significantly lower compared to those using HIR (p < 0.05), except for the LAD middle segment (p = 0.954) and LCX distal segment (p = 0.100). The SNR and CNR using IMR were significantly greater compared to those using the HIR-reformatted algorithm (p < 0.001) (Table 3). Table 4 presents the quantitative analysis of coronary atherosclerotic plaques between the two algorithms. The CT value of the plaques was significantly higher with IMR than that with HIR (p < 0.05), and the CT value of the surrounding adipose tissue was significantly lower with IMR than that with HIR (p < 0.05). Compared with HIR, the SNR and CNR between plaque and surrounding adipose tissue were significantly improved with IMR (p < 0.001). Cohen's kappa test showed that the agreement between the two observers for discerning plaque characteristics was 0.77 (Kappa = 6.73, p < 0.001).