The institutional ethics committee approved this retrospective observational study (registration number: 1910006) and waived the need for informed consent. We identified 50 patients from the clinical database who underwent stress dynamic CTP scanning for the assessment of CAD at the attending physician’s discretion between September 2017 and September 2019. We excluded patients with (1) low left ventricular ejection fraction < 20%, (2) arrhythmia, (3) greater than first-degree atrioventricular block, or (4) inappropriate CTP data for CT-derived MBF quantification. Coronary artery stenosis of ≥ 50% on CTA was considered significant and was classified based on the three major coronary vessels; patients were assessed on a per-vessel basis. The radiation dose was calculated from the dose-length product in a dose report (conversion factor = 0.014) .
Scan protocol of stress dynamic computed tomography perfusion (CTP)
Stress dynamic CTP was performed using a 320-row multi-detector CT system (Aquilion ONE GENESIS Edition, Canon Medical Systems Corporation, Otawara, Japan) as a part of the comprehensive cardiac CT protocol with a partial modification of the previous protocol . The scan timing of dynamic CTP was independently optimized, with the timing bolus scan using a 20%-diluted contrast medium, to set at 6 s before the arrival of the contrast medium at the ascending aorta. Contrast medium (iopamidol, 370 mg iodine/mL; Bayer Yakuhin, Ltd., Osaka, Japan) and a saline chaser were administered at the same injection rate and volume as the timing bolus scan, 3 min after adenosine triphosphate loading (0.16 mg/kg/min). The stress dynamic CTP dataset was obtained using the prospective electrocardiogram-gated dynamic mode, targeting a phase of 45% of the RR interval. The scan parameters for CTP were as follows: tube voltage, 80 kVp; tube current, 300 mA; gantry rotation speed, 0.275 s/rotation; detector collimation, 320 × 0.50 mm; and effective coverage, 100 mm. Subsequently, coronary CTA and delayed-enhancement CT were performed at 10 min and 15 min after stress dynamic CTP, respectively.
Post-processing of dynamic myocardial CTP images
A 360° full-reconstruction algorithm, the adaptive iterative dose reduction in three-dimensional (3D) processing (AIDR 3D, FC03, strong), and the non-rigid registration algorithm for motion compensation were used for CTP image reconstruction. The trans-axial images in the dynamic CTP dataset were reconstructed with 1.0-mm slice thickness, as the original images. In addition, three other CTP datasets were generated by adding simulated noise corresponding to the dose reduction rate (presented as tube amplitude for the dose) through a statistical model tool including both quantum noise and electronic noise. These datasets had values with 25% (225 mA), 50% (150 mA), and 75% dose reductions (75 mA) from the original dose (300 mA) . The image filtering process with 4D-SF was used after noise simulation processing (Figure 1). Finally, the four CTP datasets (original, 25%-, 50%-, and 75%-dose reduced-simulated images with 4D-SF) were evaluated using a dedicated workstation (Vitrea, Canon Medical Systems Corporation, Japan) for image quality and CT-MBF analyses.
Image quality analyses
Image quality analyses were performed for the four different CTP datasets using 5-mm-thick cardiac short-axis CTP images by average intensity projection reformat. An experienced radiologist (with 6 years of experience in cardiac imaging) selected a representative single phase (approximately 4 s after the time point of maximal enhancement in the ascending aorta) at the optimal phase for the assessment of myocardial ischemia from a series of dynamic CTP datasets, as previously described .
Regarding qualitative image quality, two radiologists (with 5 and 6 years of experience in cardiac imaging), who were blinded to all clinical and reconstruction information, independently evaluated the four different CTP datasets in random order in terms of the noise, contrast, and contour sharpness using a 5-grade scale (1, non-diagnostic; 2, fair; 3, moderate; 4, good; and 5, excellent) in the optimal window level/width settings for each case . Discrepancies between the two observers were resolved by consensus.
Regarding quantitative image quality, the other experienced radiologist (with 6 years of experience in cardiac imaging) evaluated myocardial CT attenuation (Hounsfield unit) and the standard deviation (SD) by placing the regions of interest (ROIs) (100–150 mm2) in the center of each myocardial segment based on a 16-segment model without the apex . The ROI (50–100 mm2) in the nearby skeletal muscle (latissimus dorsi, pectoralis, or intercostal) was defined as the reference tissue . The signal-to-noise ratio (SNR) was calculated by dividing the myocardial CT attenuation of an ROI by the SD of the same ROI. The contrast-to-noise ratio (CNR) was calculated by dividing the difference in CT attenuation between the myocardium and reference tissue by the SD of the reference tissue .
Computed tomography-derived myocardial blood flow analyses
Two experienced radiologists (with 5 and 6 years of experience in cardiac imaging) analyzed myocardial peak CT attenuation, time to peak (TTP), and CT-MBF in the four different dynamic CTP datasets, independent of all the information. CT-MBF was semi-automatically quantified using the Renkin-Crone equation, validated with oxygen-15-labelled water positron emission tomography . Global CT-MBF was defined as the mean of all 16 segmental values.
Continuous data are expressed as the mean (SD) or as the median (first quartile–third quartile) according to the distribution. Regarding the intra- and inter-observer agreements, Cohen kappa (k) statistics were used for the visual image quality scores, and intra-class correlation coefficients (ICC) were used for the quantitative CTP-derived quantitative parameters such as peak CT attenuation, TTP, and CT-MBF. Differences were compared between the original images and each simulated CTP image using the Dunnett test. In all tests, statistical significance was determined at p < 0.05. Statistical analyses were performed using JMP13 (SAS Institute, Cary, NC, USA).