Continuous patients who underwent calcium scoring and coronary CT angiography examinations at our hospital (People's Hospital Affiliated to Shandong First Medical University) for suspected of CAD between September 2024 and March 2025 were retrospectively collected. Inclusion criteria: (1) patients with calcium plaques. (2) over 18 years old. Exclusion criteria: motion artifacts or stent implantation. The study protocol was approved by the institutional ethics committee and written informed consent was taken from all the patients. Sample size estimation of this study was calculated by G*Power software (14).

All patients were scanned 320-row CT (Aquilion One Genesis Edition, Canon Medical Systems, Japan). Patients with heart rate > 70 beats/min were given oral beta-blockers prior to the scan. Calcium scoring was scanned with prospective ECG triggering at a 75% R-R interval. Scan parameters were following: 120 kV tube voltage, automatic tube current modulation (standard deviation 40 HU), detector collimation 320 × 0.5 mm, rotation time 0.275 s/circle, field of view 320 mm, matrix size 512 × 512. The scanning range was from the carina to the level of the diaphragm, including the entire heart. Radiation dose including the CT dose index (CTDIvol) and dose length product (DLP) of each patient were recorded. The effective dose (ED) was calculated as ED = DLP×0.014 mSv/(mGy·cm).

Calcium scoring data were reconstructed using FBP (filtered back projection, FC12), HIR [Adaptive Iterative Dose Reduction (AIDR) 3D, FC12] and DLR [Advanced Intelligent Clear-IQ Engine (AiCE), Cardiac kernel], respectively. All images were reconstructed at a slice thickness/interval of 3.0 mm. The images were transferred to post-processing software (CCSpro, Neusoft Medical) which was extensively trained on FBP and HIR reconstructed images for calcium quantification analysis. The technology first identified and extracted calcified plaques through threshold segmentation and connected component analysis, then it calculated the calcification score accordingly. Simultaneously, it segmented the cardiac tissue and constructed the vascular tree structure based on the segmentation results, which helped delineate vascular pathways. Finally, it automatically completed calcification localization and total calcification measurement by integrating the plaque extraction, cardiac tissue segmentation, and vascular tree construction results. All images were transferred to post-processing workstation (Vitrea Workstation) for image quality evaluation.

A radiologist with 7 years work experience in diagnosis of cardiac imaging drew the circle region of interest (ROI) on the aortic root, left ventricle and erector spinae muscle. The area of circle ROI was 100 mm². The mean CT value and standard deviation (SD) of ROI were recorded, signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR) was calculated by the following formula: SNRaorticroot=CTaorticrootSDaorticroot,SNRleftventricle=CTleftventricleSDleftventricle,CNRaorticroot= SCTaorticroot−CTerectorspinaemuscleSDerectorspinaemuscle,CNRleftventricle=CTleftventricle−CTerectorspinaemuscleSDerectorspinaemuscle

Two radiologists (with 10 years work experience in diagnosis of cardiac imaging) blindly evaluated subjective image quality using a 4-point scale: 1 (Poor): Non-diagnostic image quality with indistinct calcium delineation; 2 (Fair): Significant quality degradation but sufficient for diagnostic interpretation; 3 (Good): Mild quality reduction with adequate calcium visualization; 4 (Excellent): Optimal image quality with sharp calcium definition (15, 16).

Two radiologists with 3 and 8 years work experience in diagnosis of cardiac imaging performed two calculations (the minimum one week interval between two calculations) of the Agatston score, calcium volume and mass using a commercial automated calcium analysis software. The software automatically identified coronary arteries by convolutional neural network and calcified lesions by segmentation of pixels with density larger than 130 HU and area larger than 1 mm². The radiologist manually removed the calcified lesions which are not belong to the coronary arteries and recorded the results of Agatston score, calcium volume and mass. All results underwent re-review by a senior radiologist with 10 years work experience in diagnosis of cardiac imaging. Patients were classified into five risk categories: 0 (very low), 1–10 (low), 11–100 (intermediate), 101–400 (high), and >400 (very high) (17, 18).

Statistical analyses were performed with SPSS 21.0. Categorical data were reported as frequency and percentage. Continuous variables were initially analyzed by the Shapiro–Wilk test to verify the data distribution. The normal distribution data were expressed mean ± SD and one-way ANOVA test with Bonferroni correction was performed, while non-normal distribution data were reported as median (Q₁, Q₃) and Kruskal–Wallis test were performed. The consistency of the inter-observer image quality score was assessed by Cohen's Kappa test. The Agatston-based risk reclassification is evaluated by weigh Kappa. The degree of agreement was as follows: 0.00–0.20, 0.21–0.40, 0.41–0.60, 0.61–0.80, or 0.81–1.00, indicating slight, fair, moderate, substantial, or almost perfect agreement, respectively. The intra- and inter-observer reproducibility of Agatston score was assessed by intraclass correlation efficient (ICC). P < 0.05 was considered statistically significant.

The sample size estimation of this study was 105. 110 patients were initially collected according to inclusion criteria. Five patients with motion artifacts and one patient with stent implantation were excluded (Figure 1). Finally, 104 patients (63 male, 41 female; mean age 63.32 ± 7.05 years) (Table 1) were included for analysis.

DLR demonstrated significantly lower image noise and higher SNR and CNR compared to HIR and FBP (all p < 0.05, Table 2). HIR showed significantly lower CT value of aortic root and left ventricle (p < 0.05, Table 2), however CT value of aortic root and left ventricle of DLR were comparable with those of FBP and HIR (all p > 0.05, Table 2).

Two radiologists showed excellent interobserver agreement (kappa = 0.94) in assessment of subjective image quality. DLR received significantly higher subjective scores (3.80 ± 0.40) than HIR (3.48 ± 0.50) and FBP (2.36 ± 0.48) (both p < 0.001).

The calcium quantification results were presented in Table 3. The Agatston score, calcium volume and mass showed no significant difference among the three reconstruction algorithms (p = 0.155, p = 0.446 and p = 0.069, respectively). The intra- and inter-observer reproducibility of Agatston score were good in three reconstruction algorithms (ICC of FBP > 0.85, ICC of AIDR > 0.90, ICC of AiCE > 0.95, Talbe 4). Compared with FBP and DLR images, the calcified lesions were not identified in HIR images (Figure 2). In risk classification analysis, HIR reclassified one patient from moderate risk to low risk, five patients from high risk to moderate risk, and six patients from very high risk to high risk. In addition, DLR reclassified two patients from high risk to moderate risk and three patients from very high risk to high risk (Table 5). The weighted Kappa showed a value of 0.90 (95% CI, 0.85–0.96) between HIR and FBP, and 0.96 (95% CI, 0.92–0.99) between DLR and FBP, indicating better agreement between DLR and FBP in risk classification.