Advertisement

The effect of scan and patient parameters on the diagnostic performance of AI for detecting coronary stenosis on coronary CT angiography

Open AccessPublished:February 03, 2022DOI:https://doi.org/10.1016/j.clinimag.2022.01.016

      Highlights

      • AI-QCT identifies ≥50% stenosis with high sensitivity, specificity and accuracy compared to QCA.
      • AI-QCT results are unaffected by scan technique, radiation dose, contrast, or patient parameters.
      • AI-QCT is an expedient, noninvasive tool for accurately identifying moderate stenosis.

      Abstract

      Objectives

      To determine whether coronary computed tomography angiography (CCTA) scanning, scan preparation, contrast, and patient based parameters influence the diagnostic performance of an artificial intelligence (AI) based analysis software for identifying coronary lesions with ≥50% stenosis.

      Background

      CCTA is a noninvasive imaging modality that provides diagnostic and prognostic benefit to patients with coronary artery disease (CAD). The use of AI enabled quantitative CCTA (AI-QCT) analysis software enhances our diagnostic and prognostic ability, however, it is currently unclear whether software performance is influenced by CCTA scanning parameters.

      Methods

      CCTA and quantitative coronary CT (QCT) data from 303 stable patients (64 ± 10 years, 71% male) from the derivation arm of the CREDENCE Trial were retrospectively analyzed using an FDA-cleared cloud-based software that performs AI-enabled coronary segmentation, lumen and vessel wall determination, plaque quantification and characterization, and stenosis determination. The algorithm's diagnostic performance measures (sensitivity, specificity, and accuracy) for detecting coronary lesions of ≥50% stenosis were determined based on concordance with QCA measurements and subsequently compared across scanning parameters (including scanner vendor, model, single vs dual source, tube voltage, dose length product, gating technique, timing method), scan preparation technique (use of beta blocker, use and dose of nitroglycerin), contrast administration parameters (contrast type, infusion rate, iodine concentration, contrast volume) and patient parameters (heart rate and BMI).

      Results

      Within the patient cohort, 13% demonstrated ≥50% stenosis in 3 vessel territories, 21% in 2 vessel territories, 35% in 1 vessel territory while 32% had <50% stenosis in all vessel territories evaluated by QCA. Average AI analysis time was 10.3 ± 2.7 min. On a per vessel basis, there were significant differences only in sensitivity for ≥50% stenosis based on contrast type (iso-osmolar 70.0% vs non isoosmolar 92.1% p = 0.0345) and iodine concentration (<350 mg/ml 70.0%, 350-369 mg/ml 90.0%, 370–400 mg/ml 90.0%, >400 mg/ml 95.2%; p = 0.0287) in the context of low injection flow rates. On a per patient basis there were no significant differences in AI diagnostic performance measures across all measured scanner, scan technique, patient preparation, contrast, and individual patient parameters.

      Conclusion

      The diagnostic performance of AI-QCT analysis software for detecting moderate to high grade stenosis are unaffected by commonly used CCTA scanning parameters and across a range of common scanning, scanner, contrast and patient variables.

      Condensed abstract

      An AI-enabled quantitative CCTA (AI-QCT) analysis software has been validated as an effective tool for the identification, quantification and characterization of coronary plaque and stenosis through comparison to blinded expert readers and quantitative coronary angiography. However, it is unclear whether CCTA screening parameters related to scanner parameters, scan technique, contrast volume and rate, radiation dose, or a patient's BMI or heart rate at time of scan affect the software's diagnostic measures for detection of moderate to high grade stenosis. AI performance measures were unaffected across a broad range of commonly encountered scanner, patient preparation, scan technique, intravenous contrast and patient parameters.

      Abbreviations:

      AI (Artificial Intelligence), AI-QCT (Artificial Intelligence Enabled Quantitative Coronary Computed Tomographic Angiography), CAD (Coronary Artery Disease), CAD-RADS (Coronary Artery Disease Reporting and Data System), CCTA (Coronary computed tomography angiography), DS (Dual source), FBP (Filtered Back Projection), FDA (Food and Drug Administration), FFR (Fractional Flow Reserve), IR (Iterative Reconstruction), LAD (Left Anterior Descending), LCx (Left Circumflex), LM (Left Main), MDCT (Multidetector Computed Tomography), PDA (Posterior Descending Artery), QCA (quantitative coronary angiography), RCA (Right Coronary Artery), SS (Single source)

      Keywords

      1. Background

      Coronary computed tomography angiography (CCTA) has established itself as an effective tool for diagnosing and grading coronary artery disease (CAD) severity through high performance against the current gold standard, invasive coronary angiography (ICA).
      • Miller J.M.
      • et al.
      Diagnostic performance of coronary angiography by 64-row CT.
      • Budoff M.J.
      • et al.
      Diagnostic performance of 64-multidetector row coronary computed tomographic angiography for evaluation of coronary artery stenosis in individuals without known coronary artery disease: results from the prospective multicenter ACCURACY (Assessment by coronary computed tomographic angiography of individuals undergoing invasive coronary Angiography) trial.
      Furthermore, data derived through CCTA improves clinical care by guiding disease prognostication, reducing unnecessary invasive testing and improving overall outcomes.
      • Min J.K.
      • et al.
      Prognostic value of multidetector coronary computed tomographic angiography for prediction of all-cause mortality.
      • Hadamitzky M.
      • et al.
      Prognostic value of coronary computed tomographic angiography for prediction of cardiac events in patients with suspected coronary artery disease.
      • Chang H.J.
      • et al.
      Selective referral using CCTA versus direct referral for individuals referred to invasive coronary angiography for suspected CAD: a randomized, controlled, open-label trial.
      • Investigators S.-H.
      • et al.
      Coronary CT angiography and 5-year risk of myocardial infarction.
      While these scans provide data that have direct clinical utility, they also provide a quantity of data that qualitative reads by human readers cannot generate in a clinically useful period of time.
      The use of artificial intelligence (AI) for image analysis and interpretation has further enhanced our diagnostic and prognostic capability by using pixel level analytics and automated learning algorithms for rapid adaptation.
      • Motwani M.
      • et al.
      Machine learning for prediction of all-cause mortality in patients with suspected coronary artery disease: a 5-year multicentre prospective registry analysis.
      • van Rosendael A.R.
      • et al.
      Maximization of the usage of coronary CTA derived plaque information using a machine learning based algorithm to improve risk stratification; insights from the CONFIRM registry.
      • Ambale-Venkatesh B.
      • et al.
      Cardiovascular event prediction by machine learning: the multi-ethnic study of atherosclerosis.
      Given AI guided CCTA's (AI-QCT's) consistency, expediency and high performance for stenosis grading against gold standards like expert readers, quantitative invasive coronary angiography (QCA) and fractional flow reserve (FFR), CCTA analysis is likely to employ AI on a widespread basis in the future.
      • Choi A.D.
      • et al.
      CT EvaLuation by ARtificial Intelligence For Atherosclerosis, Stenosis and Vascular MorphologY (CLARIFY): a multi-center, international study.
      • Griffin W.F.
      • et al.
      AI evaluation of coronary stenosis on CT coronary angiography, comparison with quantitative coronary angiography and fractional flow reserve; a CREDENCE trial sub-study.
      While AI-QCT may have the capability to rapidly and accurately analyze CCTA images, widespread use will require an understanding of its performance across the various parameter settings. At this early stage, it is not yet clear whether AI-QCT's performance changes across common scanning parameters. However, if AI-QCT performance is comparable to QCA for plaque analysis without limitation across scan parameters, this would support the widespread use of a safe, noninvasive mode of mild to moderate lesion diagnosis and further AI-QCT generated atherosclerosis analyses on a grand scale.
      This study evaluated the sensitivity, specificity and accuracy of AI-QCT analysis compared to QCA for identifying moderate to high grade stenosis (≥50% stenosis) across a range of commonly encountered scanner, patient preparation, scan technique, intravenous contrast and patient parameters.

      2. Methods

      2.1 Subjects

      We retrospectively evaluated data from 303 patients including CCTA, FFR and QCA from the derivation arm of the Computed TomogRaphic Evaluation of Atherosclerotic Determinants of Myocardial IsChEmia (CREDENCE) Trial.
      • Rizvi A.
      • et al.
      Rationale and design of the CREDENCE trial: computed TomogRaphic evaluation of atherosclerotic DEtermiNants of myocardial IsChEmia.
      • Stuijfzand W.J.
      • et al.
      Stress myocardial perfusion imaging vs coronary computed tomographic angiography for diagnosis of invasive vessel-specific coronary physiology: predictive modeling results from the computed tomographic evaluation of atherosclerotic determinants of myocardial ischemia (CREDENCE) trial.
      The CREDENCE trial (clinicaltrials.gov NCT02173275) was a prospective, multicenter diagnostic derivation-validation, controlled clinical trial recruiting patients from 2014 to 2017.
      • Rizvi A.
      • et al.
      Rationale and design of the CREDENCE trial: computed TomogRaphic evaluation of atherosclerotic DEtermiNants of myocardial IsChEmia.
      • Stuijfzand W.J.
      • et al.
      Stress myocardial perfusion imaging vs coronary computed tomographic angiography for diagnosis of invasive vessel-specific coronary physiology: predictive modeling results from the computed tomographic evaluation of atherosclerotic determinants of myocardial ischemia (CREDENCE) trial.
      Sites and Investigators are listed in Appendix B. All enrolled CREDENCE subjects underwent CCTA, and QCA with FFR. The IRB of each site approved the study protocol and patients provided written informed consent. Inclusion and exclusion criteria and clinical sites are listed in the supplement (Appendix A). This study is an investigator-initiated study. Cleerly Inc. had no role in study design or performance. Cleerly Inc. performed the CCTA analyses for the study in a blinded manner and provided statistical services as determined and requested by the study investigators.

      2.2 CT imaging protocols

      CCTA was performed using a CT scanner with ≥64-detector rows. Sites were instructed to perform CCTA in accordance with guidelines from the Society of Cardiovascular Computed Tomography (SCCT).
      • Rizvi A.
      • et al.
      Rationale and design of the CREDENCE trial: computed TomogRaphic evaluation of atherosclerotic DEtermiNants of myocardial IsChEmia.
      • Leipsic J.
      • et al.
      SCCT guidelines for the interpretation and reporting of coronary CT angiography: a report of the Society of Cardiovascular Computed Tomography Guidelines Committee.

      2.3 CCTA scanning parameters

      The software's sensitivity, specificity and accuracy for detecting coronary lesions ≥50% stenosis were tested across a multitude of screening parameters including scanner vendor (GE, Philips, Siemens, Toshiba), scanner model, single vs. dual source scanning, gating technique (prospective vs. retrospective helical), bolus type (tracker vs. test bolus) and injection rate, contrast agent (Iodixanol [Visipaque] vs. all others), contrast volume and iodine concentration, radiation parameters including tube voltage and dose length product, mode of image generation (filter back projection vs. iterative reconstruction), administration of beta blockers or nitroglycerine prior to scanning, and patient's average heart rate and BMI at time of scan. The AI-software's diagnostic performance measures were determined based on concordance with QCA measurements and subsequently compared across screening parameters.

      2.4 Quantitative coronary angiography

      Invasive coronary angiography was performed in agreement with clinical indications and imaging standards. A dedicated core laboratory performed blinded QCA in two orthogonal views on a per lesion basis of every lesion visually ≥30% diameter stenosis in vessels with a reference vessel diameter ≥2.0 mm. Lesions estimated to be less than 30% were recorded as no stenosis.

      2.5 Artificial intelligence-based segmentation and stenosis quantification

      The AI-based approach to CCTA interpretation in this study was performed using an FDA-cleared software service (Cleerly Lab, Cleerly, New York, NY) that performs automated analysis of CCTA using a series of validated convolutional neural network models (including VGG19 network, 3D U-Net, and VGG Network Variant) for image quality assessment, coronary segmentation and labeling, lumen wall evaluation and vessel contour determination (Fig. 1) and plaque characterization.
      United States Food and Drug Administration
      Cleerly Labs 510 (k) premarket notification.
      Training and testing were performed on a proprietary database. A centerline algorithm was developed from 1,007,945 images, which incorporated 23,068 vessels from 3671 patients. Lumen and vessel wall algorithms were developed from 1,414,877 images, comprised of 8555 vessels from 3676 patients. First, the AI-aided approach produces a centerline along each coronary artery for lumen and outer vessel wall contouring. This is applied to each phase of the examination. The two optimal series are then identified for further analysis. These top two phases are evaluated interactively on a per vessel basis, e.g., the right coronary artery (RCA) will be reconstructed from the phase which yields the highest RCA image quality, while the posterior descending artery (PDA) may come from the second phase if the PDA has a higher image quality on that phase.
      Fig. 1
      Fig. 1A stepwise explanation of how AI-QCT calculates % diameter stenosis. AI-QCT segmentation and analysis of a CCTA from a 53 yr old male with chest pain and a positive nuclear stress test. A. Initially, all available series are evaluated by two machine learned algorithms to select the two series with optimal image quality for further analysis. B. A centerline algorithm is performed along each coronary artery. C. Once coronary artery segmentation is performed, an automated labeling (blue line and boxes) is done to classify arteries by their location as well the proximal, mid and distal portions within a single vessel. D and E. The lumen (purple) and outer vessel wall (yellow) contouring boundaries are determined. F. For each plaque, the cross-sectional slice that demonstrates the greatest absolute narrowing, % diameter stenosis severity and remodeling index is automatically calculated (73% proximal LAD stenosis with a RI of 1.7 depicted). G. A color overlay is performed based upon HU attenuation, in this case a predominantly noncalcified plaque is present with a low-density non-calcified core (red overlay). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
      Once coronary artery segmentation is performed, segments are labeled in an automated fashion to classify the segment by vessel, and whether it is in the proximal, mid or distal portion of that vessel. For stenosis evaluation the software places markers at the beginning and end of each lesion and selects the mean coronary diameter at the closest normal proximal reference cross section as the reference diameter (Dref), and the mean diameter on the cross section demonstrating the greatest absolute stenosis (Ds). The % diameter stenosis is then automatically calculated using the following formula: % Diameter stenosis = (1 – (Ds/Dref)) x 100. After the AI algorithm has finished all operations, as mandated by the FDA, a quality control cardiac CT trained technician reviews the results and makes manual adjustments if necessary. AI computational analysis time was 10.3 ± 2.7 min.

      2.6 Stenosis comparison QCA vs CCTA

      QCA stenosis was measured by core lab readers in each vessel territory. If QCA detected stenosis, AI-based stenosis evaluation was then conducted on those coronary segments using an SCCT 18-segment coronary tree model. When QCA evaluated a vessel territory and recorded 0% stenosis, that value was applied to each segment in the territory. However, if QCA was present in only one segment of the territory, and identified stenoses, values were treated as missing in the other segments. Thus, the denominator for per-segment analysis equals the sum of all segments in all territories where QCA is 0% and all segments where QCA was measured and recorded as >0%. The maximum QCA and AI-based diameter stenosis were calculated across segments for each stenotic vessel. Where QCA was measured in just one segment in the vessel, that value was applied to the vessel. Per-vessel territory and per-patient results were obtained in a similar manner. No cases were excluded due to impaired image quality on CCTA. If impaired image quality was present due to motion, beam hardening or other artifact, just the portion of the coronary artery with poor quality was excluded from the analysis. Any quantitative data from the excluded segment is not included in the final report.

      2.7 Statistical analysis

      Analysis was performed using SAS software version 9.4 (SAS Inc., Cary, NC). The diagnostic performance of AI-based diameter stenosis was evaluated by calculating the sensitivity, specificity and diagnostic accuracy for identifying ≥50% stenosis on a segment, vessel, vessel territory and patient basis using QCA as the reference standard. Area under the receiver operating characteristic curves was used to evaluate the diagnostic performance for ≥50% stenosis per QCA. The comparability of the continuous measures of diameter stenosis for AI and QCA were assessed via correlation using Pearson's correlation coefficients and calculation of the mean difference (bias). Sensitivity, specificity, and diagnostic performance for ≥50% diameter stenosis per vessel territory were compared across patient subgroups using logistic generalized estimating equation (GEE) regression models, to account for the potential correlation of multiple vessel territories per patient.
      AI diagnostic performance on a per patient basis was compared across categorical scan parameters using the Chi-square or Fisher's exact test. Ordered parameters were compared using the Jonckheere-Terpstra test. Diagnostic performance measures for vessel territories were compared across scan parameters using a logistic generalized estimating equation model to account for any within subject correlation of the three vessel territories measured per subject.

      3. Results

      3.1 Study population

      Baseline characteristics of the study population are listed in Table 1. A total of 303 of the 307 (98%) subjects comprising the derivation arm of the CREDENCE cohort
      • Rizvi A.
      • et al.
      Rationale and design of the CREDENCE trial: computed TomogRaphic evaluation of atherosclerotic DEtermiNants of myocardial IsChEmia.
      • Stuijfzand W.J.
      • et al.
      Stress myocardial perfusion imaging vs coronary computed tomographic angiography for diagnosis of invasive vessel-specific coronary physiology: predictive modeling results from the computed tomographic evaluation of atherosclerotic determinants of myocardial ischemia (CREDENCE) trial.
      were included; 4 patients were excluded due to corruption of their CT imaging data. Of the 303 patients evaluated, 175 (57.8%) had obstructive disease while 128 (42.2%) had non-obstructive disease.
      Table 1Patient characteristics.
      Patient parameters
      Age64 ± 10
      Male71% (218)
      BMI26 ± 4
      Race
       Black2% (7)
       Asian71% (217)
       White27% (82)
      Hypertension64% (197)
      Hyperlipidemia44% (136)
      Diabetes31% (95)
      Current Smoker17% (53)
      Prior Smoker34% (103)
      # Diseased Vessel Territories (≥50% CCTA stenosis)
       032% (96)
       135% (105)
       221% (64)
       313% (38)
      Scan parameters
      Scanner vendor
       General Electric18% (54)
       Philips2% (5)
       Siemens43% (132)
       Toshiba38% (116)
      Tube Voltage
       70 kV0.3% (1)
       80 kV6% (17)
       100 kV38% (116)
       120 kV52% (160)
       Other4% (11)
      Gating technique
       Prospective/sequential31% (95)
       Retrospective helical61% (188)
       Single beat acquisition8% (23)

      3.2 AI analysis success

      The AI algorithm was run successfully in all 303 patients (100%). There was a total of 171,195 mm of vessel length evaluated in the entire cohort, of this, a total of 1861 mm (1.09%) was excluded due to impaired image quality. The length of an exclusion averaged 14.1 (± 13.9) mm and were longer in the RCA (15.22 ± 6.72 mm) than the LAD (7.62 ± 4.12 mm) or the Circumflex (7.58 ± 8.13 mm) arteries.

      3.3 AI diagnostic measurements by scanning parameters at the patient level

      AI sensitivity, specificity, and accuracy measurements for detecting moderate stenosis at the patient level are listed for each of the tested scanning parameters in Table 2. The table reflects that at the patient level, there were no significant differences in AI diagnostic measurements based on scanner type (vendor, model, single vs. dual source), gating technique, bolus type, contrast features (agent, volume, iodine concentration), radiation features (tube voltage, dose length product), image generation (filtered back projection vs. iterative reconstruction), medication administration prior to imaging (beta blockade, nitroglycerine) or clinical (Fig. 2, Fig. 3) patient features at time of scan (heart rate, BMI). While the p values for specificity of scanner vendor and scanner model were 0.13 and 0.2, respectively, these values are low suggesting a trend, but do not meet statistical significance (P < 0.05).
      Table 2AI Diagnostic performance by scanning parameter on the patient level.
      SubgroupSensitivitySpecificityAccuracy
      Scanner vendor
       GE96% (24/25)82.8% (24/29)89.9% (48/54)
       Philips100% (3/3)50% (1/2)80.0% (4/5)
       Siemens94.7% (71/75)60% (33/55)80.0% (104/130)
       Toshiba93.2% (69/74)70% (28/40)85.1% (97/114)
      0.92270.13120.4194
      Scanner model
       Toshiba – Aquilion One93.2% (69/74)70.0% (28/40)85.1% (97/114)
       GE-VCT Light Speed 6496.0% (24/25)84.0% (21/25)90.0% (45/50)
      Siemens Dual Source Definition100% (25/25)58.3% (14/24)76.6% (39/49)
       All others92.5% (49/53)62.2% (23/37)80% (72/90)
      0.70290.20010.3787
      Single source vs dual source
       SS93.8% (137/146)71.3% (67/94)85% (204/240)
       DS96.8% (30/31)59.4% (19/32)77.8% (49/63)
      >0.9990.21160.1693
      Average heart rate during CT
       ≤70 bpm94.6% (139/147)69.0% (69/100)84.2% (208/247)
       >70 bpm95.2% (20/21)68.8% (11/16)83.8% (31/37)
      >0.999>0.9990.9471
       ≤75 bpm94.9% (149/157)69.4% (75/108)84.5% (224/265)
       >75 bpm90.9% (10/11)62.5% (5/8)79.0% (15/19)
      0.46480.70200.5165
       ≤7094.6% (139/147)69.0% (69/100)84.2% (208/247)
       71–80100% (16/16)69.2% (9/13)86.2% (25/29)
       >8080.0% (4/5)66.7% (2/3)75.0% (6/8)
      0.97550.98010.9123
      Beta blocker IV
       Yes94.0% (141/150)67.3% (76/113)82.5% (217/263)
       No96.3% (26/27)76.9% (10/13)90.0% (36/40)
      >0.9990.47830.2344
      Nitroglycerin dose
       None95.9% (47/49)65.9% (29/44)81.7% (76/93)
       Low (0.1–0.4 mg)95.2% (40/42)65.7% (23/35)81.8% (63/77)
       High (>0.4 mg)93.0% (80/86)72.3% (34/47)85.7% (114/133)
      0.45690.50540.3941
       None95.9% (47/49)65.9% (29/44)81.7% (76/93)
       Some93.8% (120/128)69.5% (57/82)84.3% (177/210)
      0.72870.67870.5790
      Bolus tracker or test bolus
       BT95.3% (101/106)67.2% (43/64)84.7% (144,185)
       TB93.0% (66/71)70.5% (43/61)82.6% (109/132)
      0.52490.69020.6185
      Gating technique
       Prospective/sequential/Single Beat91.9% (68/74)72.1% (31/43)84.6% (99/117)
       Retrospective helical96.1% (99/103)67.1% (55/82)83.2% (154/185)
      0.32400.68530.7527
      Contrast agent
       Visipaque100% (5/5)77.8% (7/9)85.7% (12/14)
       All others94.2% (162/172)67.5% (79/117)83.4% (241/289)
      >0.9990.7179>0.999
      Concentration mg iodine per mL
       Q1: <350100% (5/5)80.0% (8/10)86.7% (13/15)
       Q2: 350–36994.1% (32/34)70.3% (26/37)81.7% (58/71)
       Q3: 370–40090.8% (59/65)65.2% (30/46)80.2% (89/111)
       Q4: >40097.3% (71/73)68.8% (22/32)88.6% (93.105)
      0.37310.59120.2411
      Contrast injection rate (cc/sec)
       <589.5% (34/38)66.7% (20/30)79.4% (54/68)
       595.3% (102/107)72.7% (56/77)85.9% (158/184)
       >596.9% (31/32)52.9% (9/17)81.6% (40/49)
      0.16590.55540.6059
      Contrast volume- mg/kg body weight
       Q1: <333.489.5% (34/38)71.1% (27/38)80.3% (61/76)
       Q2: 333.4–408.293.3% (42/45)63.3% (19/30)81.3% (61/75)
       Q3: 408.3–504.598.0% (49/50)68.0% (17/25)88.0% (66/75)
       Q4: >504.595.5% (42/44)71.9% (23/32)85.5% (65/76)
      0.16880.91000.2378
      Tube voltage
       ≤80 kV100% (8/8)100% (10/10)100% (18/18)
       100 kV91.8% (67/73)60.0% (24/40)80.5% (91/113)
       120 kV95.4% (83/87)70.8% (51/72)84.3% (134/159)
      0.56890.85290.8520
      Dose length product (mGy*cm)
       Q1: <200.591.3% (42/46)64.3% (18/28)81.1% (60/74)
       Q2: 200.5–36595.2% (40/42)71.9% (23/32)85.1% (63/74)
       Q3: 366–600.298.0% (42/43)71.0% (22/31)86.5% (64/74)
       Q4: >600.293.2% (41/44)68.8% (22/32)82.9% (63/76)
      0.59300.77090.7296
      BMI
       <3094.4% (151/160)69.6% (71/102)84.7% (222/262)
       ≥3094.1% (16/17)62.5% (15/24)75.6% (31/41)
      >0.9990.50090.1434
      Iterative reconstruction
       FBP96.0% (24/25)84.0% (21/25)90.0% (45/50)
       IR94.1% (143/152)64.4% (65/101)82.2% (208/253)
      Scanner vendor>0.9990.05890.1753
      Fig. 2
      Fig. 2Comparison of performance in two scans of differing image quality secondary to BMI differences. CCTA of a 53-year-old woman with a BMI of 34 demonstrates increased noise and low signal-to-noise ratio. Curved multiplanar reformat (A) depicts scattered nonobstructive plaques. AI-QCT analysis (B) depicts scattered mixed plaques and a 38% stenosis in the mid vessel; QCA depicted a 44% stenosis in this location. CCTA on a 62-year-old male with BMI 25 has significantly improved overall IQ and SNR (C). A 76% stenosis was depicted on AI-QCT, which corresponded to a 70% stenosis determined on QCA.
      Fig. 3
      Fig. 3Image quality secondary to iodine concentration differences with conventional coronary angiography correlation. CCTA on a 57 year old man using iodine with a concentration of 400 mg iodine per mL demonstrates high intra-arterial attenuation (480 HU) (A). AI-QCT (B) depicts a 50% stenosis in the mid LAD (arrow) which as determined to be 56% on quantitative coronary angiography (QCA) (arrow, C). CCTA on a 65 year old woman using iodine with a concentration of 320 mg iodine per mL demonstrates decreased intra-arterial attenuation (210 HU) (D). A 46% stenosis was depicted on AI QCT (arrow S), which corresponded to a 45% stenosis on QCA (arrow, E).

      3.4 AI diagnostic measurements by scanning parameters at the level of the vessel territory

      AI sensitivity, specificity, and accuracy measurements for detecting moderate stenosis at the level of the vessel territory, and per patient, are listed for each of the tested scanning parameters in Table 3. Data indicate that there were significant differences in sensitivity for detecting 50% stenosis based on contrast type (iso-osmolar 70.0% vs non isoosmolar 92.1% p = 0.0345) and iodine concentration (<350 mg/ml 70.0%, 350-369 mg/ml 90.0%, 370–400 mg/ml 90.0%, >400 mg/ml 95.2%; p = 0.0287). There was no significant difference in the specificity or accuracy which remained comparable across both iodine concentrations and contrast types. There were no significant differences observed in diagnostic performance measurements for any other scanning parameters.
      Table 3AI diagnostic performance by scanning parameter at the level of the vessel territory.
      SubgroupSensitivitySpecificityAccuracy
      Scanner vendor
       GE86.1% (37/43)87.4% (104/119)87.0% (141/162)
       Philips100% (6/6)77.8% (7/9)86.7% (13/15)
       Siemens92.2% (95/103)84.0% (241/287)86.2% (336/390)
       Toshiba92.0% (103/112)81.3% (187/230)84.8% (290/342)
      0.62390.64730.8013
      Scanner model
       Toshiba – Aquilion One92.0% (103/112)81.3% (187/230)84.8% (290/342)
       GE-VCT Light Speed 6486.1% (37/43)86.9% (93/107)86.7% (130/150)
       Siemens Dual Source Definition96.9% (31/32)83.5% (96/115)86.4% (127/147)
       All others92.5% (49/53)87.5% (98/112)89.1% (147/165)
      0.34140.47760.6559
      Single source vs dual source
       SS91.0% (202/222)83.5% (416/498)85.8% (618/720)
       DS92.9% (39/42)83.7% (123/147)85.7% (162/189)
      0.67790.88150.9693
      Average heart rate during CT
       ≤70 bpm90.9% (200/220)83.7% (436/521)85.8% (636/741)
       >70 bpm96.9% (31/32)82.3% (65/79)86.4% (96/111)
      0.30360.88780.8863
       ≤75 bpm92.9% (13/14)83.3% (464/557)85.8% (682/795)
       >75 bpm91.6% (218/238)86.1% (37/43)87.7% (50/57)
      0.88650.61190.7284
       ≤7090.9% (200/220)83.7% (436/521)85.8% (636/741)
       71–80100% (25/25)80.7% (50/62)86.2% (75/87)
       >8085.7% (6/7)88.2% (15/17)87.5% (21/24)
      0.64200.97180.8392
      Beta blocker IV
       Yes91.3% (42/46)81.1% (60/74)85.0% (102/120)
       No91.3% (199/218)83.9% (479/571)85.9% (678/789)
      0.99100.54130.8010
      Nitroglycerin dose
       None91.9% (57/62)83.4% (181/217)85.3% (238/279)
       Low (0.1–0.4 mg)89.7% (61/68)80.4% (131/163)83.1% (192/231)
       High (>0.4 mg)91.8% (123/134)85.7% (227/265)87.7% (350/399)
      0.99150.49260.3450
       None91.9% (57/62)83.4% (181/217)85.3% (238/279)
       Some91.1% (184/202)83.6% (358/428)86.0% (542/630)
      0.79140.99380.7855
      Bolus tracker or test bolus
       BT92.4% (146/158)81.5% (287/352)84.9% (433/510)
       TB89.6% (95/106)86.2% (250/290)87.1% (345/396)
      0.45690.14920.3681
      Gating technique
       Prospective/sequential/single beat87.9% (102/116)83.0% (195/235)84.6% (297/351)
       Retrospective helical93.9% (139/148)84.0% (342/407)86.7% (481/555)
      0.09380.70010.4257
      Contrast agent
       Visipaque70.0% (7/10)87.5% (28/32)83.3% (35/42)
       All others92.1% (234/254)83.4% (511/613)85.9% (745/867)
      0.03450.56240.6777
      Concentration mg iodine per mL
       Q1: <35070% (7/10)88.6% (31/35)84.4% (38/45)
       Q2: 350–36990% (45/50)85.3% (139/163)86.4% (184/213)
       Q3: 370–40090.0% (90/100)79.8% (186/233)82.9% (276/333)
       Q4: >40095.2% (99/104)85.8% (181/211)88.9% (280/315)
      0.02870.99200.3223
      Contrast injection rate (cc/sec)
       <588.7% (55/62)82.4% (117/142)84.3% (172/204)
       591.7% (143/156)85.9% (340/396)87.5% (483/552)
       >593.5% (43/46)76.2% (77/101)81.6% (120/147)
      0.34560.47410.6803
      Contrast volume- mg/kg body weight
       Q1: <333.487.5% (49/56)84.3% (145/172)85.1% (194/228)
       Q2: 333.4–408.290.8% (59/65)80.0% (128/160)83.1% (187/225)
       Q3: 408.3–504.593.3% (70/75)84.7% (127/150)87.6% (197/225)
       Q4: >504.592.6% (63/68)85.6% (137/160)87.7% (200/228)
      0.26710.55180.2633
      Tube voltage
       ≤80 kV91.7% (11/12)90.5% (38/42)90.7% (49/54)
       100 kV88.6% (101/114)80.9% (182/225)83.5% (283/339)
       120 kV93.6% (117/125)84.7% (298/352)87.0% (415/477)
      0.22570.78810.6652
      Dose length product (mGy*cm)
       Q1: <200.587.8% (42/46)84.5% (125/148)85.6% (190/222)
       Q2: 200.5–36591.8% (40/42)88.2% (142/161)89.2% (198/222)
       Q3: 366–600.295.1% (42/43)77.0% (124/161)82.0% (182/222)
       Q4: >600.290.9% (41/44)84.6% (137/162)86.4% (197/228)
      0.43240.32950.6740
      BMI
       <3091.6% (219/239)83.9% (459/547)86.3% (678/786)
       ≥3088.0% (22/25)81.6% (80/98)82.9% (102/123)
      0.62320.56860.3194
      Iterative reconstruction
       FBP86.1% (37/43)86.9% (93/107)86.7% (130/150)
       IR92.3% (204/221)82.9% (446/538)85.6% (650/759)
      0.20730.49850.7591

      4. Discussion

      This study evaluated the effect of CCTA scanning parameters on the performance of an FDA-cleared AI-enabled software capable of identifying and quantifying coronary stenoses and found that on a per-patient basis the software's sensitivity, specificity and accuracy for identifying lesions with ≥50% stenosis were unaffected by scanner type, gating technique, contrast features, radiation features, mode of image reconstruction or patients' clinical features at time of scan.
      The natural variability amongst readers, the sheer volume of data as well as the time expenditure required for each read provide a foundation for the introduction of AI guided CCTA into the field of cardiac imaging.
      • Choi A.D.
      • et al.
      CT EvaLuation by ARtificial Intelligence For Atherosclerosis, Stenosis and Vascular MorphologY (CLARIFY): a multi-center, international study.
      To test the software's performance for stenosis identification and for plaque characterization, the CLARIFY trials compared the AI software against three gold standards, expert readers, QCA and FFR and showed AI performed strongly for stenosis identification and better than expert readers for plaque morphology.
      • Choi A.D.
      • et al.
      CT EvaLuation by ARtificial Intelligence For Atherosclerosis, Stenosis and Vascular MorphologY (CLARIFY): a multi-center, international study.
      • Griffin W.F.
      • et al.
      AI evaluation of coronary stenosis on CT coronary angiography, comparison with quantitative coronary angiography and fractional flow reserve; a CREDENCE trial sub-study.
      This raises the question of whether accuracy and performance favors AI in addition to improving read expediency and data processing. Further studies are currently underway to evaluate this AI software's capability for plaque quantification and characterization against intravascular ultrasound and optical coherence tomography. However, based on current data, there is a strong basis for the use of AI for the identification or exclusion of moderate to severe stenoses given high performance against QCA, experts, and FFR.
      As AI becomes more widely accepted in CCTA analysis, it becomes necessary to test its performance across a variety of scanning parameters as widespread use will require it to perform well across all parameter variants. But additionally, the functional and anatomical variability that necessitates adjustments in scanning parameters further reinforces the importance of high AI performance across a multitude of different settings. To date, AI has been used to minimize motion artifact in circumstances where patients are nonresponsive to beta blockers,
      • Fuchs T.A.
      • et al.
      Impact of a new motion-correction algorithm on image quality of low-dose coronary CT angiography in patients with insufficient heart rate control.
      and to optimize phase selection when patients have abnormal heart rhythms.
      • Stassi D.
      • et al.
      Automated selection of the optimal cardiac phase for single-beat coronary CT angiography reconstruction.
      • Oda S.
      • et al.
      A low tube voltage technique reduces the radiation dose at retrospective ECG-gated cardiac computed tomography for anatomical and functional analyses.
      • Lee A.M.
      • et al.
      Coronary computed tomography angiography during arrhythmia: radiation dose reduction with prospectively ECG-triggered axial and retrospectively ECG-gated helical 128-slice dual-source CT.
      While these programs paved the way for AI use for optimization of the imaging process, the AI software discussed in this study attacks the problem of interpretation itself, not just image optimization. Uniquely, this software was designed to automatically identify and calculate stenoses based on a series of algorithms that segment the coronary tree, identify a midline around which perpendicular slices are compared to identify the beginning, end and maximum point of stenosis in a lesion. The program subsequently employs deep learning to enhance its performance as it is exposed to more vessels. This is the first study to analyze the application of this AI program for stenosis identification across scanning parameters and supports that the software's performance is on par with the performance of QCA, an invasive gold standard for stenosis identification.
      • Griffin W.F.
      • et al.
      AI evaluation of coronary stenosis on CT coronary angiography, comparison with quantitative coronary angiography and fractional flow reserve; a CREDENCE trial sub-study.
      CCTA images are generated using a variety of different scanning parameter combinations purposefully chosen by the CT technologist to optimize image quality while minimizing radiation exposure. While various societal guidelines recommend specific techniques, investigators in the PROTECT VI (Prospective Multicenter Registry On RadiaTion Dose Estimates Of Cardiac CT AngIOgraphy IN Daily Practice in 2017) found that the protocols varied widely and that recommended techniques were not routinely adhered to.
      • Stocker T.J.
      • et al.
      Application of low tube potentials in CCTA: results from the PROTECTION VI study.
      Likely this was a function of the added necessity and complexity of adjusting parameters based on the clinical characteristics of patients at the time of scan.
      • Husmann L.
      • et al.
      Body physique and heart rate variability determine the occurrence of stair-step artefacts in 64-slice CT coronary angiography with prospective ECG-triggering.
      In routinely performed clinical CCTA exams, selected scanning protocol parameters are selected based on patient requirements to produce diagnostically readable, though sometimes varied appearing images.
      For CCTA, the gating technique has one of the greatest impacts on image appearance, diagnostic quality and radiation dose. Prospective gating is commonly used with lower heart rates. This technique minimizes radiation exposure by using a “step-and-shoot” approach of limited exposure during a small part of the cardiac cycle, usually diastole. The alternative is to continuously scan throughout the cardiac cycle, known as known as retrospective helical gating.
      • Earls J.P.
      How to use a prospective gated technique for cardiac CT.
      Prospective gating has been shown to have higher image quality scores than retrospective gating,
      • Earls J.P.
      • et al.
      Prospectively gated transverse coronary CT angiography versus retrospectively gated helical technique: improved image quality and reduced radiation dose.
      however, these results, and others from similar studies were dependent on patient specific variables including low and consistent heart rates.
      • Husmann L.
      • et al.
      Body physique and heart rate variability determine the occurrence of stair-step artefacts in 64-slice CT coronary angiography with prospective ECG-triggering.
      • Stolzmann P.
      • et al.
      Dual-source CT in step-and-shoot mode: noninvasive coronary angiography with low radiation dose.
      • Muenzel D.
      • et al.
      Step and shoot coronary CT angiography using 256-slice CT: effect of heart rate and heart rate variability on image quality.
      • Yang L.
      • et al.
      64-MDCT coronary angiography of patients with atrial fibrillation: influence of heart rate on image quality and efficacy in evaluation of coronary artery disease.
      Despite the potential for significant image quality differences between the techniques, we did not find that the gating technique affected diagnostic performance of the algorithm.
      The image reconstruction techniques are known to affect image quality. Filtered back projection (FBP) has been the standard reconstruction algorithm since the invention of CT 50 years ago but has been known to produce images with greater noise than more recent techniques.
      • Schofield R.
      • et al.
      Image reconstruction: part 1 - understanding filtered back projection, noise and image acquisition.
      Iterative reconstruction (IR) is a newer and more computationally intense method which can typically produce images of better quality at lower radiation dose than FBP.
      • Tayal U.
      • et al.
      Image reconstruction in cardiovascular CT: part 2 - iterative reconstruction; potential and pitfalls.
      • Yin W.H.
      • et al.
      Iterative reconstruction to preserve image quality and diagnostic accuracy at reduced radiation dose in coronary CT angiography: an intraindividual comparison.
      However, the additional filters used in IR scan generation can sharpen the appearance of a structure's edges which may result in loss of fine detail.
      • Tayal U.
      • et al.
      Image reconstruction in cardiovascular CT: part 2 - iterative reconstruction; potential and pitfalls.
      Consistently, at low radiation doses the Agatston score calculated from a calcium scoring CT based on IR generated images could vary by up to 15% from those calculated based on images using FBP.
      • den Harder A.M.
      • et al.
      Submillisievert coronary calcium quantification using model-based iterative reconstruction: a within-patient analysis.
      As with gating techniques, we did not find any performance differences for the algorithm based on the reconstruction technique.
      While our overall data show that there are no significant differences in AI performance compared to QCA across the scan parameters tested, there are a few caveats regarding iodine concentration, scanner model and scanner vendor that require further discussion. With regard to differences in iodine concentrations, it is important to recognize that the contrast type, its concentration of iodine, and injection rates all contribute to lumen enhancement. While high iodine concentrations increase vessel attenuation, they also increase patient heart rate, introducing potential motion artifacts into the image
      • Honoris L.
      • et al.
      Comparison of contrast enhancement, image quality and tolerability in coronary CT angiography using 4 contrast agents: a prospective randomized trial.
      or obscure the distinction between calcified and non-calcified plaque.
      • Cademartiri F.
      • et al.
      Intravenous contrast material administration at helical 16-detector row CT coronary angiography: effect of iodine concentration on vascular attenuation.
      • Maffei E.
      • et al.
      Plaque imaging with CT coronary angiography: effect of intra-vascular attenuation on plaque type classification.
      Iso-osmolar contrasts are consequently preferable but require higher flow rates at injection to achieve similar levels of vessel attenuation as dyes containing higher iodine concentrations.
      • Honoris L.
      • et al.
      Comparison of contrast enhancement, image quality and tolerability in coronary CT angiography using 4 contrast agents: a prospective randomized trial.
      • Kim T.
      • et al.
      Effects of injection rates of contrast material on arterial phase hepatic CT.
      While our findings indicate that AI was less sensitive to stenosis on a per vessel basis with Iodixanol (Visipaque) use, it is notable that the patients who received this dye were injected at a non-standard flow rate of 2–3 mL/s, well below the standard practice guidelines of 4–6 mL/s, likely resulting in low vessel attenuation. Additionally, the patient level specificity across scanner model and vendor were equivocal, likely due to low power of our cohort. However, overall, these data strongly support the use of AI guided CCTA based on a strong performance against a QCA gold standard.
      This study has limitations. The present study was a post-hoc analysis of the CREDENCE trial and, while it is unexpected that significant bias would be introduced in a retrospective evaluation leveraging blinded core laboratory readers, it nevertheless emphasizes the absence of a prospective clinical trial that should be performed in the future. Further, this study evaluated the AI-based evaluation for measures of stenosis severity, rather than for plaque volume, composition, vascular remodeling and other important CAD metrics. Also, ground truth in this present study was core-lab interpreted QCA for any stenosis >30%, in keeping with prior multicenter studies employing QCA. Because of this, we are not able report the diagnostic performance of the AI-based evaluation for the presence of stenosis lower than this range. Stenoses in this range have been historically considered inconsequential by QCA, although newer data suggest a prognostic significance to these “mild” lesions that may nevertheless possess high-risk atherosclerotic characteristics.

      5. Conclusion

      AI -QCT can identify moderate to high grade stenoses with high sensitivity, specificity and accuracy compared to QCA, the invasive gold standard. Furthermore, the diagnostic performance measurements of AI are unaffected by CCTA scanner, scan technique, radiation dose, patient preparation, contrast, and patient parameters.

      Declaration of competing interest

      Equity Interest Cleerly, Inc.: ADC, JPE, HM; Employee Cleerly, JPE, RSJ, TRC.

      Appendix A. Inclusion and exclusion criteria

      Inclusion criteria

      1. Age >18 years
      2. Scheduled to undergo clinically indicated non-emergent invasive coronary angiography

      Exclusion criteria

      1. Known CAD before index testing (myocardial infarction, percutaneous coronary intervention, or coronary artery bypass graft surgery)
      2. Hemodynamic instability
      3. Inability to provide written informed consent
      4. Concomitant participation in another clinical trial in which subject is subject to investigational drug or device
      5. Pregnant state
      6. Contraindication to iodinated contrast due to prior near-fatal anaphylactic reaction (laryngospasm, bronchospasm, cardiorespiratory collapse, or equivalent)
      7. Serum creatinine ≥1.7 mg/dl or Glomerular Filtration Rate <30 ml/min
      8. Baseline irregular heart rhythm (e.g., atrial fibrillation, etc.)
      9. Heart rate ≥100 beats per minute
      10. Systolic blood pressure ≤90 mmHg
      11. Contraindications to β blockers or nitroglycerin or adenosine
      12. BMI >40 kg/m2

      Appendix B. Participating sites and enrolling investigators

      Investigator: Erick Avelar, MD
      Oconee Heart and Vascular Center at St Mary's Hospital
      Athens, GA, USA
      Investigator: Hyuk-Jae Chang, MD PhD3
      Severance Cardiovascular Hospital and Severance Biomedical Science Institute, Yonsei University College of Medicine, Yonsei University Health System Seoul, South Korea
      Investigator: Jung Hyun Choi, MD PhD
      Pusan National University Hospital Busan, South Korea
      Investigator: Jason Cole, MD
      Mobile Cardiology Associates Mobile, Alabama, USA
      Investigator: Joon-Hyung Doh, MD
      Inje University Ilsan Paik Hospital Goyang, South Korea
      Investigator: Andrejs Erglis, MD
      Pauls Stradins Clinical University Hospital Riga, Latvia
      Investigator: Alessia Gimelli, MD, Dante Chiappino, MD
      Fondazione Toscana Gabriele Monasterio Pisa, Italy
      Investigator: Ae-Young Her, MD
      Kang Won National University Hospital Chuncheon, South Korea
      Investigator: James J. Jang, MD
      Kaiser Permanente Hospital San Jose, CA, USA
      Investigator: Muhammad Akram Khan, MD, Venkaa Chilakapati, MD, Ahmed Ladak, MD, Irfan Ullah, MD
      Cardiac Center of Texas McKinney, Texas, USA
      Investigators: Paul Knaapen, MD, PhD
      Amsterdam University Medical Center VU University Medical Center Amsterdam, the Netherlands
      Investigator: Bon-Kwon Koo, MD PhD
      Seoul National University Hospital Seoul, South Korea
      Investigator: Bin Lu, MD
      State Key Laboratory of Cardiovascular Disease, Fuwai Hospital Beijing, China
      Investigator: Atizazul Mansoor, MD
      Pinnacle Health Cardiovascular Institute Harrisburg, PA, USA
      Investigator: Faisal Nabi, MD
      Houston Methodist Hospital Houston, Texas, USA
      Investigators: Ryo Nakazato, MD, Hiroyumi Niinuma, MD
      St. Luke's International Hospital Tokyo, Japan
      Investigator: Chang-Wook Nam, MD PhD
      Keimyung University Dongsan Hospital Daegu, South Korea
      Investigator: Hyung-Bok Park, MD
      International St. Mary's Hospital Catholic Kwandong University College of Medicine Incheon, South Korea
      Investigator: Michael Ridner, MD
      Heart Center Research, LLC Huntsville, Alabama, USA
      Investigator: Chris Rowan, MD
      Renown Heart and Vascular Institute Reno, NV, USA
      Investigator: U. Joseph Schoepf, MD, Daniel Sternberg, MD
      Medical University of South Carolina Charleston, SC, USA
      Investigator: Sang-Hoon Shin, MD
      National Health Insurance Service Ilsan Hospital Goyang, South Korea
      Investigator: Randall C. Thompson, MD
      St. Luke's Mid America Heart Institute Kansas City, MO, USA

      References

        • Miller J.M.
        • et al.
        Diagnostic performance of coronary angiography by 64-row CT.
        N Engl J Med. 2008; 359: 2324-2336
        • Budoff M.J.
        • et al.
        Diagnostic performance of 64-multidetector row coronary computed tomographic angiography for evaluation of coronary artery stenosis in individuals without known coronary artery disease: results from the prospective multicenter ACCURACY (Assessment by coronary computed tomographic angiography of individuals undergoing invasive coronary Angiography) trial.
        J Am Coll Cardiol. 2008; 52: 1724-1732
        • Min J.K.
        • et al.
        Prognostic value of multidetector coronary computed tomographic angiography for prediction of all-cause mortality.
        J Am Coll Cardiol. 2007; 50: 1161-1170
        • Hadamitzky M.
        • et al.
        Prognostic value of coronary computed tomographic angiography for prediction of cardiac events in patients with suspected coronary artery disease.
        JACC Cardiovasc Imaging. 2009; 2: 404-411
        • Chang H.J.
        • et al.
        Selective referral using CCTA versus direct referral for individuals referred to invasive coronary angiography for suspected CAD: a randomized, controlled, open-label trial.
        JACC Cardiovasc Imaging. 2019; 12: 1303-1312
        • Investigators S.-H.
        • et al.
        Coronary CT angiography and 5-year risk of myocardial infarction.
        N Engl J Med. 2018; 379: 924-933
        • Motwani M.
        • et al.
        Machine learning for prediction of all-cause mortality in patients with suspected coronary artery disease: a 5-year multicentre prospective registry analysis.
        Eur Heart J. 2017; 38: 500-507
        • van Rosendael A.R.
        • et al.
        Maximization of the usage of coronary CTA derived plaque information using a machine learning based algorithm to improve risk stratification; insights from the CONFIRM registry.
        J Cardiovasc Comput Tomogr. 2018; 12: 204-209
        • Ambale-Venkatesh B.
        • et al.
        Cardiovascular event prediction by machine learning: the multi-ethnic study of atherosclerosis.
        Circ Res. 2017; 121: 1092-1101
        • Choi A.D.
        • et al.
        CT EvaLuation by ARtificial Intelligence For Atherosclerosis, Stenosis and Vascular MorphologY (CLARIFY): a multi-center, international study.
        J Cardiovasc Comput Tomogr. 2021; 15: 470-476
        • Griffin W.F.
        • et al.
        AI evaluation of coronary stenosis on CT coronary angiography, comparison with quantitative coronary angiography and fractional flow reserve; a CREDENCE trial sub-study.
        J Am Coll Cardiol Img. 2022; https://doi.org/10.1016/j.jcmg.2021.10.020
        • Rizvi A.
        • et al.
        Rationale and design of the CREDENCE trial: computed TomogRaphic evaluation of atherosclerotic DEtermiNants of myocardial IsChEmia.
        BMC Cardiovasc Disord. 2016; 16: 190
        • Stuijfzand W.J.
        • et al.
        Stress myocardial perfusion imaging vs coronary computed tomographic angiography for diagnosis of invasive vessel-specific coronary physiology: predictive modeling results from the computed tomographic evaluation of atherosclerotic determinants of myocardial ischemia (CREDENCE) trial.
        JAMA Cardiol. 2020; 5: 1338-1348
        • Leipsic J.
        • et al.
        SCCT guidelines for the interpretation and reporting of coronary CT angiography: a report of the Society of Cardiovascular Computed Tomography Guidelines Committee.
        J Cardiovasc Comput Tomogr. 2014; 8: 342-358
        • United States Food and Drug Administration
        Cleerly Labs 510 (k) premarket notification.
        2019
        • Fuchs T.A.
        • et al.
        Impact of a new motion-correction algorithm on image quality of low-dose coronary CT angiography in patients with insufficient heart rate control.
        Acad Radiol. 2014; 21: 312-317
        • Stassi D.
        • et al.
        Automated selection of the optimal cardiac phase for single-beat coronary CT angiography reconstruction.
        Med Phys. 2016; 43: 324
        • Oda S.
        • et al.
        A low tube voltage technique reduces the radiation dose at retrospective ECG-gated cardiac computed tomography for anatomical and functional analyses.
        Acad Radiol. 2011; 18: 991-999
        • Lee A.M.
        • et al.
        Coronary computed tomography angiography during arrhythmia: radiation dose reduction with prospectively ECG-triggered axial and retrospectively ECG-gated helical 128-slice dual-source CT.
        J Cardiovasc Comput Tomogr. 2012; 6: 172-183 e2
        • Stocker T.J.
        • et al.
        Application of low tube potentials in CCTA: results from the PROTECTION VI study.
        JACC Cardiovasc Imaging. 2020; 13: 425-434
        • Husmann L.
        • et al.
        Body physique and heart rate variability determine the occurrence of stair-step artefacts in 64-slice CT coronary angiography with prospective ECG-triggering.
        Eur Radiol. 2009; 19: 1698-1703
        • Earls J.P.
        How to use a prospective gated technique for cardiac CT.
        J Cardiovasc Comput Tomogr. 2009; 3: 45-51
        • Earls J.P.
        • et al.
        Prospectively gated transverse coronary CT angiography versus retrospectively gated helical technique: improved image quality and reduced radiation dose.
        Radiology. 2008; 246: 742-753
        • Stolzmann P.
        • et al.
        Dual-source CT in step-and-shoot mode: noninvasive coronary angiography with low radiation dose.
        Radiology. 2008; 249: 71-80
        • Muenzel D.
        • et al.
        Step and shoot coronary CT angiography using 256-slice CT: effect of heart rate and heart rate variability on image quality.
        Eur Radiol. 2011; 21: 2277-2284
        • Yang L.
        • et al.
        64-MDCT coronary angiography of patients with atrial fibrillation: influence of heart rate on image quality and efficacy in evaluation of coronary artery disease.
        Am J Roentgenol. 2009; 193: 795-801
        • Schofield R.
        • et al.
        Image reconstruction: part 1 - understanding filtered back projection, noise and image acquisition.
        J Cardiovasc Comput Tomogr. 2020; 14: 219-225
        • Tayal U.
        • et al.
        Image reconstruction in cardiovascular CT: part 2 - iterative reconstruction; potential and pitfalls.
        J Cardiovasc Comput Tomogr. 2019; 13: 3-10
        • Yin W.H.
        • et al.
        Iterative reconstruction to preserve image quality and diagnostic accuracy at reduced radiation dose in coronary CT angiography: an intraindividual comparison.
        JACC Cardiovasc Imaging. 2013; 6: 1239-1249
        • den Harder A.M.
        • et al.
        Submillisievert coronary calcium quantification using model-based iterative reconstruction: a within-patient analysis.
        Eur J Radiol. 2016; 85: 2152-2159
        • Honoris L.
        • et al.
        Comparison of contrast enhancement, image quality and tolerability in coronary CT angiography using 4 contrast agents: a prospective randomized trial.
        Int J Cardiol. 2015; 186: 126-128
        • Cademartiri F.
        • et al.
        Intravenous contrast material administration at helical 16-detector row CT coronary angiography: effect of iodine concentration on vascular attenuation.
        Radiology. 2005; 236: 661-665
        • Maffei E.
        • et al.
        Plaque imaging with CT coronary angiography: effect of intra-vascular attenuation on plaque type classification.
        World J Radiol. 2012; 4: 265-272
        • Kim T.
        • et al.
        Effects of injection rates of contrast material on arterial phase hepatic CT.
        Am J Roentgenol. 1998; 171: 429-432