Author + information
- Received January 17, 2018
- Revision received July 8, 2018
- Accepted July 12, 2018
- Published online July 1, 2019.
- Csilla Celeng, MD, PhDa,∗ (, )
- Tim Leiner, MD, PhDa,
- Pál Maurovich-Horvat, MD, PhD, MPHb,
- Béla Merkely, MD, PhDb,
- Pim de Jong, MD, PhDa,
- Jan W. Dankbaar, MD, PhDa,
- Hendrik W. van Es, MD, PhDc,
- Brian B. Ghoshhajra, MD, MBAd,
- Udo Hoffmann, MD, MPHd and
- Richard A.P. Takx, MD, PhD, MSca,c,d
- aDepartment of Radiology, University Medical Center Utrecht, Utrecht, the Netherlands
- bMTA-SE Cardiovascular Imaging Research Group Heart and Vascular Center, Semmelweis University, Budapest, Hungary
- cDepartment of Radiology, St. Antonius Hospital, Nieuwegein, the Netherlands
- dCardiac MR PET CT Program, Department of Radiology, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts
- ↵∗Address for correspondence:
Dr. Csilla Celeng, Department of Radiology, University Medical Center Utrecht, Heidelberglaan 100, 3584 CX Utrecht, the Netherlands.
Objectives This meta-analysis determined the diagnostic performance of coronary computed tomography (CT) angiography (CTA), CT myocardial perfusion (CTP), fractional flow reserve CT (FFRCT), the transluminal attenuation gradient (TAG), and their combined use with CTA versus FFR as a reference standard for detection of hemodynamically significant coronary artery disease (CAD).
Background CTA provides excellent anatomic, albeit limited functional information for the evaluation of CAD. Recently, various functional CT techniques emerged to assess the hemodynamic consequences of CAD.
Methods This meta-analysis was performed in adherence to the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. PubMed, EMBASE, and Web of Science were searched from inception until September 7, 2017. Bayesian random effects analysis was used to compute pooled sensitivity, specificity, and the summary receiver-operating characteristic curve of the index tests and compare them with the FFR as a reference standard. Analyses were performed on vessel and patient levels. Because CTA has excellent sensitivity, specificity was considered most relevant. Individual FFRCT values were collected.
Results Overall, 54 articles and 5,330 patients were included. At vessel level, pooled specificity of CTP (0.86; 95% confidence interval [CI]: 0.76 to 0.93), FFRCT (0.78; 95% CI: 0.72 to 0.83) and TAG (0.77; 95% CI: 0.61 to 0.89) were substantially higher than that of CTA (0.61; 95% CI: 0.54 to 0.68). The addition of FFRCT, CTP, and TAG to CTA resulted in high to excellent specificities (0.80 to 0.92). The summary receiver-operating characteristic curve at vessel level yielded superior diagnostic accuracy for CTP, FFRCT, and combined CTA and CTP, compared with CTA. A subanalysis of on-site versus off-site FFRCT revealed no substantial differences between the sensitivity (0.84 vs. 0.85) and specificity (0.80 vs. 0.73) of the 2 techniques. In a second subanalysis, dynamic CTP showed higher sensitivity (0.85 vs. 0.72), but had a lower specificity (0.81 vs. 0.90) than static CTP.
Conclusions CTP and FFRCT demonstrated a substantial improvement in the identification of hemodynamically significant CAD compared with CTA; therefore, their integration to clinical workflow before revascularization is recommended.
- coronary artery disease
- computed tomography
- fractional flow reserve
- myocardial perfusion imaging
Computed tomography angiography (CTA) has a high sensitivity in the evaluation of coronary artery disease (CAD) (1–4). However, its specificity for the detection of hemodynamically significant CAD is still moderate at best, especially in cases of intermediate coronary stenosis (30% to 70% reduction in coronary lumen diameter) (5).
Invasive fractional flow reserve (FFR) is the current reference standard for determining the hemodynamic significance of CAD (6–8). FFR is calculated as the ratio between the maximum achievable myocardial blood flow during hyperemia in the presence of a stenosis and the maximum achievable myocardial blood flow in the absence of a stenosis (6). FFR has a “grey zone” of ischemic threshold value (between 0.75 and 0.80) (7). Previous comprehensive trials showed that the use of FFR-guided revascularization reduced the rate of composite endpoints (death, nonfatal myocardial infarction, repeat revascularization) and urgent revascularization (9,10). Computed tomography myocardial perfusion imaging (CTP) has been proposed as an alternative to magnetic resonance imaging (MRI) and nuclear cardiology techniques for the diagnosis of hemodynamically significant CAD (11). In addition, new noninvasive CT techniques have emerged, which allow the assessment of potential hemodynamic changes attributable to CAD without modification of imaging protocols or an additional radiation dose. Advances in computational fluid dynamics allow for determination of significant ischemia by computation of noninvasive FFR CT (FFRCT) or the transluminal attenuation gradient (TAG) (12–15). The aim of this study was to conduct a meta-analysis to determine the diagnostic performance of CTA, CTP, FFRCT, and TAG, and the combined use of these techniques with CTA versus invasive FFR as the reference standard for the detection of hemodynamically significant CAD in patients with suspected or known CAD.
This systematic review and meta-analysis was conducted in agreement with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement (16). The protocol was published online (PROSPERO 2015 CRD42015017148). PubMed, EMBASE, and Web of Science were systematically searched from inception to September 7, 2017. The search syntax is presented in Supplemental Table 1. No search restrictions were imposed for language or date of publication. A manual reference check of included articles was performed to identify additional articles missed by our systematic search.
Two independent researchers (C.C. and R.A.P.T.) applied pre-defined criteria to determine article eligibility. Articles were included based on the following criteria: study domain–patients with suspected or known CAD; index test–CTA, CTP, FFRCT, and TAG, and the combination of CTA and CTP, CTA, and FFRCT, and CTA and TAG; reference standard–invasive FFR for intermediate coronary lesions; study result–agreement between the index and reference standard; and study design–cross-sectional study. In cases of studies with overlapping populations, the study with the larger population sample was included. Discordances between reviewers were resolved by consensus. Animal studies, phantom studies, and case reports (n < 10) were not included.
Data extraction and critical appraisal
Data were extracted by one reviewer (C.C.) and verified by a second reviewer (R.A.P.T.). The following data categories were extracted from the included studies: study population characteristics, index test characteristics, and reference test characteristics. True positive, false positive, false negative, and true negative numbers were obtained. The results were summarized in 2-by-2 contingency tables. In cases of combined modalities, the reviewers included only those studies that judged the combined test positive if both CTA and functional CT were positive for hemodynamically significant CAD. This means that patients with a positive test on CTA but negative on functional CT, or a negative CTA and positive test on functional CT, were deemed as not having hemodynamically significant CAD. In cases of multiple imaging thresholds, the pre-defined primary endpoint was selected. If the primary endpoint was not reported, the best performing threshold was chosen. In addition, the performance of FFRCT was evaluated at vessel level by digitizing scatterplots or Bland-Altman plots, in which invasive FFR was used as the reference standard using a previously described method by Cook et al. (17). Because individual points in a scatterplot can overlap, studies with >10% missing data points were omitted from the analysis. If individual FFRCT values were listed in a table, then those values were extracted. A threshold of invasive FFR of ≤0.80 was used as the cutoff for hemodynamically significant CAD. Two independent reviewers (C.C. and R.A.P.T.) assessed the relevance and validity of the included studies using the QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies) criteria (18). Discrepancies were resolved by consensus. The maximum time between noninvasive and invasive stenosis degree assessment was also recorded. If this period was >2 months, the flow and timing of the study was scored as high bias.
Data synthesis and analysis
Analyses were performed at both vessel and patient level. Possible publication bias per imaging technique was assessed graphically by generating funnel plots for modalities that included ≥10 studies. Sensitivity, specificity, positive likelihood ratio (+LR) and negative LR (−LR), and the log diagnostic odds ratio, including the 95% confidence interval, were calculated. Pooled results were used to calculate the summary receiver-operating curves (SROCs). Heterogeneity was assessed by visual inspection of SROCs (19). Due to anticipated methodological heterogeneity, Bayesian random effects analysis was performed, because it provides more robust results in cases of studies with a relatively small sample size. Furthermore, it also takes the potential sources of variation (e.g., imprecision of sensitivity and specificity) into account (20–23). The logit link function was used. The penalized complexity prior framework was applied, which resulted in more precise estimates (23). The contrast was chosen at p = 0.05 (sigma >3), which corresponded to sensitivities or specificities within an interval of 0.5 to 0.95, with a 95% probability (22). Post-test probabilities per imaging technique were calculated based on various pre-test probabilities and the pooled LRs. Subgroup analyses were performed to reveal potential performance differences between on-site FFRCT versus off-site FFRCT and dynamic CTP versus static CTP. For the subgroup analysis of individual FFRCT at vessel level, a graph was generated that displayed the sensitivity and specificity at different FFRCT values. Youden’s index was used to determine the optimal threshold (i.e., by identification of the threshold that maximizes the sum of sensitivity plus specificity). In addition, the FFRCT value was determined, at which sensitivity or specificity was ≥90%. Agreement between invasive FFR and FFRCT values was considered when both tests were ≤0.80 or >0.80. All statistical analyses were performed with the statistical software R (version 3.42, R Foundation for Statistical Computing, Vienna, Austria) and the package “meta4diag” (version 2.0.6, arXiv:1512.06220v2 [stat.AP]) and MedCalc (version 18.5, Mariakerke, Belgium).
After removal of duplicates, the search retrieved 1,377 potentially relevant articles. Full text review was performed in 150 articles that met our pre-defined inclusion criteria. Included studies were published between 2008 and 2017. From 54 studies (n = 5,330), 39 studies reported on CTA, 12 reported on CTP, 10 assessed CTA and CTP, 6 reported on TAG, 4 reported on CTA and TAG, 18 assessed FFRCT, and 3 reported on CTA and FFRCT (Figure 1). The average weighted radiation exposure was 5.0 mSv for prospectively gated CTA, 11.4 mSv for retrospectively gated CTA, and 6.9 mSv for stress CTP only (not including rest CTA). Funnel plots did not reveal obvious publication bias (Supplemental Figure 1). Study population characteristics, index test characteristics, and reference test characteristics were retrieved and are listed in Supplemental Tables 2 to 4. Quality assessment using QUADAS-2 criteria showed that patient selection specifically was a source of bias (Figure 2).
Pooled sensitivity and specificity at vessel and patient level
At vessel level, pooled sensitivity was the best for CTA (0.87), followed by FFRCT (0.85), CTA and CTP (0.82), CTP (0.81), CTA and FFRCT (0.76), CTA and TAG (0.70), and finally TAG (0.59). The highest pooled specificity was reached by CTA and TAG (0.92), followed by CTA and CTP (0.88), CTP (0.86), CTA and FFRCT (0.80), FFRCT (0.78), TAG (0.77), and CTA (0.61). More detailed diagnostic performances are listed in Table 1 and Supplemental Figure 2. The subanalysis of on-site (11 studies) versus off-site (7 studies) FFRCT on vessel level revealed no substantial differences between the sensitivity (0.84 vs. 0.85) and specificity (0.80 vs. 0.73, respectively) of the 2 techniques (Supplemental Figure 3). The second subanalysis on dynamic CTP (7 studies) versus static CTP (5 studies) showed that dynamic CTP had higher sensitivity (0.85 vs. 0.72) but lower specificity (0.81 vs. 0.90) compared with static CTP (Supplemental Figure 4).
At patient level, pooled sensitivity was the best for CTA (0.94), followed by FFRCT (0.89), CTA and CTP (0.89), CTP (0.83), and TAG (0.69). The highest pooled specificity was reached by CTA and CTP (0.81), followed by CTP (0.79), FFRCT (0.76), CTA (0.48), and TAG (0.39). More detailed diagnostic performances are listed in Table 2 and Supplemental Figure 5.
Pooled diagnostic accuracy
The SROC at vessel level demonstrated superior diagnostic accuracy for CTP, FFRCT, and CTA in combination with CTP, compared with CTA. CTA in combination with FFRCT, TAG, and CTA in combination with TAG were associated with wide confidence intervals due to the limited number of studies; hence, the true diagnostic performance was indeterminate (Figure 3A). In line with these findings, a large change in post-test probability was observed for functional CT over CTA alone, whereas TAG alone demonstrated only a modest change in post-test probability (Figure 4). Specifically, at a pre-test probability of 30%, in case of a negative test, post-test probabilities were between 8% and 11% for CTA, CTP, FFRCT, and the combination of CTA and CTP as well as CTA and FFRCT. In case of a positive result, CTA yielded a post-test probability of 49%, whereas CTP achieved 73% and FFRCT achieved 62%. On a patient level, the SROCs showed similar good diagnostic performance for CTP, CTA in combination with CTP, and FFRCT compared with CTA (Figure 3B). TAG had a lower diagnostic performance than CTA, with wide confidence intervals.
Diagnostic accuracy of individual FFRCT at vessel level
We extracted data on 1,370 vessels from 14 studies (Supplemental Table 4) and found an optimal FFRCT threshold of ≤0.80 using the Youden index, which resulted in an area under the curve of 0.87, with a sensitivity of 84% (95% CI: 81% to 87%) and a specificity of 77% (95% CI: 74% to 80%). Using pre-determined thresholds, we observed ≥90% sensitivity at a FFRCT value of >0.82 and ≥90% specificity at a FFRCT value of ≤0.74 (Figure 5A). In vessels with FFRCT values between 0.74 and 0.82, diagnostic agreement with invasive FFR was weak (54% in 316 vessels), whereas diagnostic agreement for FFRCT values outside of this range was high (87% in 1,054 vessels). The distribution of the data also presented as the frequency of invasive FFR at a different FFRCT value (Figure 5B). The agreement between invasive FFR and FFRCT is displayed in Figure 5C.
This meta-analysis demonstrated the additional value of functional CT over anatomic CTA alone for the assessment of hemodynamically significant CAD. The main results of this investigation can be summarized as follows. 1) In line with previous observations, CTA showed excellent sensitivity; however, it had limited specificity both on vessel and patient levels. 2) Regarding the functional CT imaging techniques, CTP and FFRCT yielded high to excellent sensitivity. Interestingly, the specificity of both these imaging techniques, especially CTP, was high. Evidence for TAG was limited and showed moderate sensitivity on both levels, whereas its specificity showed substantial differences in performance at vessel and patient levels (0.77 vs. 0.39). Also, discrepancies in TAG could be dependent on the acquisition technique and scanner hardware, which might have resulted in differences in coronary contrast concentrations and thus, could also have influenced TAG results (24). 3) With the combination of CTA and CTP as well as CTA and FFRCT, vessel-level sensitivity remained high, whereas CTA and TAG showed moderate sensitivity. The specificity of all the combined techniques was high to excellent. 4) Regarding the diagnostic accuracy on both vessel- and patient-level all CT functional imaging techniques except TAG showed better performance than CTA alone; however, on vessel-level the combination of CTA and FFRCT as well as CTA and TAG showed wide confidence intervals. Therefore, their true diagnostic accuracy was unclear. 5) Concerning the changes in the diagnostic probability, at a pre-test probability of 30%, a negative test resulted in post-test probabilities of 8% to 11% for CTA, CTP, FFRCT, and the combination of CTA and CTP as well as CTA and FFRCT. Hence, a negative test decreased the absolute probability of disease by ∼20%, although no improvement was observed over CTA alone. In cases of a positive result, CTA yielded a post-test probability of 49%, whereas CTP achieved 73% and FFRCT achieved 62%; thus, these techniques more accurately identified hemodynamically significant CAD. This meta-analysis demonstrated that the use of functional CT imaging techniques might aid the clinical management of patients with suspected or known CAD.
Advantages and limitations of CTP
A previous meta-analysis demonstrated that the performance of CTP was similar to MRI and positron emission tomography for the detection of hemodynamically significant CAD (25). In the present study, which included twice the number of vessels, we showed that CTP yielded the same specificity. Despite these advantages, the integration of CTP into clinical workflow is limited. A possible limitation, which might hamper the use of CTP, is the complexity of the imaging procedure. Currently, there are 3 acquisition methods for the detection and evaluation of myocardial perfusion defects: static single energy (myocardial contrast enhancement is obtained at the early arterial phase); static dual energy (iodine material distribution map is created by using dual-energy acquisition); and dynamic CTP (myocardial contrast enhancement is acquired several times during the first-pass) (26). The subanalysis of dynamic CTP versus static CTP demonstrated that the sensitivity of dynamic CTP was higher, but its specificity was lower than that of static CTP. Both static and dynamic CTP can be performed during rest and/or stress.
Based on our findings, most institutions use a stress perfusion protocol with dynamic acquisition (53%) to improve the detection of perfusion defects (27,28). In terms of radiation exposure, static rest CTP—in line with coronary CTA—requires a single-snapshot for the detection of contrast attenuation and is therefore associated with a lower radiation dose. We found that the weighted radiation exposure of stress CTP plus rest CTA leads to substantially higher radiation (11.9 mSv) compared with that of prospectively electrocardiographically triggered CTA alone (5.0 mSv). With use of low-kilovolt protocols, dynamic CTP is feasible at a lower radiation dose (29,30). Nevertheless, radiation exposure should be limited whenever feasible, and this might entail a preference for FFRCT.
Added value of FFRCT
Our meta-analysis showed that the diagnostic performance of FFRCT was similar to that of CTP; moreover, it did not require additional radiation nor the administration of adenosine. Furthermore, it allows to estimate functional changes in coronary artery flow after stent implantation (31). Recently, Cook et al. (17) performed a systematic review, in which they included 5 FFRCT studies and digitized plots to acquire individual correlations of FFRCT and invasive FFR values. They concluded that FFRCT around the cutpoint of 0.80 showed less certainty, and that FFRCT values of >0.90 and ≤0.60 provided almost complete certainty. In our meta-analysis, we performed a similar subanalysis on a larger number of vessels (1,370 vs. 908) and observed good agreement with invasive FFR for FFRCT values >0.82 and ≤0.74. Of note, when evaluating values close to a threshold (i.e., intermediate values), lower agreement will be observed, which is also true for repeated measures of invasive FFR itself (32). Some discrepancy between modeling and directly measured FFR is expected to be present, although with advanced FFRCT algorithms, the use of machine learning, and improvements in CT hardware, this discrepancy is expected to decrease (33,34). HeartFlow (Redwood City, California) FFRCT analysis was the first commercially available software that was able to compute a CT-derived FFR value. In recent years, various vendors developed other FFRCT platforms. When using HeartFlow, CTA data are transmitted to an off-site location and are analyzed by an independent laboratory. In case of on-site assessment, FFRCT data are analyzed at the local institution. Within the frame of this meta-analysis, we performed a subanalysis to determine the diagnostic accuracy of off-site FFRCT versus on-site FFRCT. Our results indicated that the performance of on-site FFRCT algorithms was as high as that of off-site FFRCT. Of note, on-site FFRCT post-processing might be desirable because it allows the physician to access functional data in a more timely manner. Nevertheless, most FFRCT algorithms are proprietary, and the exact mechanism remains unclear.
Costs attributable to functional CT imaging
Because health care costs of noninvasive versus invasive imaging are substantially different, several studies aimed to assess the cost-effectiveness of novel noninvasive imaging techniques for identification of patients with significant CAD. The clinical and economic outcomes of FFRCT versus usual care were investigated within the framework of 3 comprehensive trials (DISCOVER-FLOW [Diagnosis of ISChemia-Causing Stenoses Obtained Via NoninvasivE FRactional FLOW Reserve], HeartFlowNXT [HeartFlow analysis of coronary blood flow using CT angiography: NeXt sTeps] and PLATFORM [Prospective LongitudinAl Trial of FFRct: Outcome and Resource IMpacts]) (35–37). Beyond improved or equivalent clinical outcome (12% to 19% fewer events at 1 year), the use of FFRCT resulted in reduced costs (30% to 33% lower) by more accurately identifying patients for revascularization. Based on these promising results, HeartFlow Inc. received Food and Drug Administration clearance in November 2014. In the United Kingdom, a study demonstrated that the integration of HeartFlow FFRCT as a diagnostic strategy for stable chest pain resulted in cost savings of £200 per patient (38), which led the National Institute for Health and Care Excellence to recommend the adoption of HeartFlow FFRCT into current practice (39). The National Institute for Health and Care Excellence considered HeartFlow FFRCT cost-effective at a list price of £700 (excluding VAT) per analysis compared with the current treatment pathway of all functional imaging tests, which includes single-photon emission computed tomography, MRI, and echocardiography. In case of TAG, no studies were reported on the additional cost of TAG software, because the analysis can be performed on-site using a dedicated workstation, thus the costs are expected to be limited. The cost-effectiveness of CTP was evaluated only in 1 study, which reported lower costs and an additional gain in quality-adjusted life years when using dual-energy CT compared with single-photon emission computed tomography (40).
The results of our meta-analysis need to be considered in the context of the included studies, which were mainly performed at expert centers and limited in number for some modalities. CT radiation exposure was not reported in all included studies. Verification bias was an important limitation, which occurred when studies only performed the reference standard in those with a positive initial test. This might result in a high pre-test probability, which could increase the specificity of a given imaging modality. Another limitation was the heterogeneity among studies (including differences in diagnostic criteria), which might have influenced the reported summary values of the different imaging modalities. Finally, we were not able to obtain individual FFRCT values from all studies, including the NXT trial (41), because only 312 individual points could be identified in the Bland-Altman plot of the 484 vessels included in the study. For the DeFACTO (Determination of Fractional Flow Reserve by Anatomic Computed Tomographic AngiOgraphy) data, a substudy (42) was used in which 150 vessels (407 in total in the original study population) of intermediate stenosis on CTA were evaluated.
This meta-analysis indicated that despite recent developments in CT hardware, the specificity of CTA remains moderate for the detection of hemodynamically significant stenosis. The specificity of functional CT imaging tests and their combined use with CTA was high to excellent, except for TAG, which showed discrepancies in specificity on vessel and patient levels. CTP, FFRCT, and its combined use with CTA yielded higher diagnostic performance than CTA alone. The diagnostic performance of individual FFRCT at vessel level was good, with a relative narrow area of uncertainty around the threshold of 0.80. CTP and FFRCT both have the potential to improve identification of patients with hemodynamically significant CAD at reduced health costs, therefore, their integration into the routine clinical workflow is recommended. Due to limited evidence the exact role of TAG is uncertain.
COMPETENCY IN MEDICAL KNOWLEDGE 1: Current clinical guidelines do not encompass the application of functional CT imaging techniques for the detection of hemodynamically relevant ischemia in patients with stable CAD.
COMPETENCY IN MEDICAL KNOWLEDGE 2: CTP and FFRCT provide improved diagnostic accuracy compared with coronary CTA alone, which has high sensitivity but moderate specificity.
COMPETENCY IN PATIENT CARE AND PROCEDURAL SKILLS: Both functional CT imaging techniques might refine the critical pathway of patients to invasive catheterization.
COMPETENCY IN INTERPERSONAL AND COMMUNICATION SKILLS: It is relevant to assemble robust evidence to clarify the role of functional CT techniques and to discuss the available options with patients who will undergo coronary CTA.
TRANSLATIONAL OUTLOOK: Further clinical trials using sophisticated imaging protocols and machine-learning algorithms are needed to validate the novel functional CT tests in the clinical setting and to further refine the probability of potential outcomes of patients with CAD.
Dr. Celeng has received grant support from the European Association of Cardiovascular Imaging. Dr. Ghoshhajra has been a consultant for Siemens Healthcare and Medtronic. Dr. Hoffman has received research grants from KOWA, MedImmune, Pfizer, Siemens, and HeartFlow, Inc. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
- Abbreviations and Acronyms
- computed tomography angiography
- coronary artery disease
- computed tomography myocardial perfusion
- fractional flow reserve
- likelihood ratio
- magnetic resonance imaging
- fractional flow reserve computed tomography
- summary receiver-operating curve
- transluminal attenuation gradient
- Received January 17, 2018.
- Revision received July 8, 2018.
- Accepted July 12, 2018.
- 2019 American College of Cardiology Foundation
- Budoff M.J.,
- Dowe D.,
- Jollis J.G.,
- et al.
- Budoff M.J.,
- Nakazato R.,
- Mancini G.B.,
- et al.
- Rossi A.,
- Papadopoulou S.L.,
- Pugliese F.,
- et al.
- De Bruyne B.,
- Sarma J.
- Jeremias A.,
- Kirtane A.J.,
- Stone G.W.
- Lubbers M.,
- Coenen A.,
- Kofflard M.,
- et al.
- Choi J.H.,
- Koo B.K.,
- Yoon Y.E.,
- et al.
- Lu M.T.,
- Ferencik M.,
- Roberts R.S.,
- et al.
- Liberati A.,
- Altman D.G.,
- Tetzlaff J.,
- et al.
- Cook C.M.,
- Petraco R.,
- Shun-Shin M.J.,
- et al.
- Guo J.,
- Riebler A.,
- Rue H.
- Takx R.A.,
- Blomberg B.A.,
- El Aidi H.,
- et al.
- Ko S.M.,
- Park J.H.,
- Hwang H.K.,
- Song M.G.
- Kim S.M.,
- Kim Y.N.,
- Choe Y.H.
- Kim K.H.,
- Doh J.H.,
- Koo B.K.,
- et al.
- Petraco R.,
- Sen S.,
- Nijjer S.,
- et al.
- Zreik M.,
- Lessmann N.,
- van Hamersvelt R.W.,
- et al.
- Douglas P.S.,
- De Bruyne B.,
- Pontone G.,
- et al.
- ↵National Institute for Health and Care Excellence. HeartFlow FFRCT for estimating fractional flow reserve from coronary CT angiography. Available at: nice.org.uk/guidance/mtg32. Accessed September 2, 2017.
- Norgaard B.L.,
- Leipsic J.,
- Gaur S.,
- et al.
- Nakazato R.,
- Park H.B.,
- Berman D.S.,
- et al.