Diagnostic Accuracy of Artificial Intelligence-Based Angiography-Derived Fractional Flow Reserve Using Pressure Wire-Based Fractional Flow Reserve as a Reference

Abstract

Background: Angiographic fractional flow reserve (angioFFR) is a novel artificial intelligence (AI)-based angiography-derived fractional flow reserve (FFR) application. We investigated the diagnostic accuracy of angioFFR to detect hemodynamically relevant coronary artery disease.

Methods and Results: Consecutive patients with 30–90% angiographic stenoses and invasive FFR measurements were included in this prospective, single-center study conducted between November 2018 and February 2020. Diagnostic accuracy was assessed using invasive FFR as the reference standard. In patients undergoing percutaneous coronary intervention, gradients of invasive FFR and angioFFR in the pre-senting segments were compared. We assessed 253 vessels (200 patients). The accuracy of angioFFR was 87.7% (95% confidence interval [CI] 83.1–91.5%), with a sensitivity of 76.8% (95% CI 67.1–84.9%), specificity of 94.3% (95% CI 89.5–97.4%), and area under the curve of 0.90 (95% CI 0.86–0.93%). AngioFFR was well correlated with invasive FFR (r=0.76; 95% CI 0.71–0.81; P<0.001). The agreement was 0.003 (limits of agreement: −0.13, 0.14). The FFR gradients of angioFFR and invasive FFR were comparable (n=51; mean [±SD] 0.22±0.10 vs. 0.22±0.11, respectively; P=0.87).

Conclusions: AI-based angioFFR showed good diagnostic accuracy for detecting hemodynamically relevant stenosis using invasive FFR as the reference standard. The gradients of invasive FFR and angioFFR in the pre-stenting segments were comparable.

Fractional flow reserve (FFR) is a physiological index used to guide decision making for coronary revascularization in patients with coronary artery disease (CAD).^1,2 Randomized clinical trials have shown that percutaneous coronary intervention (PCI) for coronary stenoses with an FFR ≤0.80 improves patient outcomes compared with medical therapy, and the use of FFR for CAD is classified as a Class Ia recommendation in clinical guidelines.^1–4 Despite these recommendations, FFR dissemination is not widespread. The main reasons for this are that invasive FFR requires the maneuvering pressure wires and pharmacology to induce maximum hyperemia, which has potential side effects.⁵

The advent of several angiography-derived FFR estimation methods has generated new opportunities to determine functional coronary relevance that do not have risks related to the pressure wire maneuvers and do not require maximum hyperemia.^6–9 The non-commercial prototype software Angiographic FFR (angioFFR; Siemens Healthcare GmbH, Forchheim, Germany), which is not for diagnostic use, is one of the first artificial intelligence (AI)-based angiography-derived FFR applications. AngioFFR uses an upfront-trained machine learning model with >27,000 synthetic geometries with angioFFR values that are based on 3-dimensional (3D) computational fluid dynamics (CFD; Figures 1,2).¹⁰ In a clinical setting, angioFFR values can then be extracted from the upfront-trained robust data without complex 3D CFD analysis. However, clinical validation studies of angioFFR compared with invasive FFR are still lacking.

Figure 1.

Schematic flow diagram of machine learning-based angiographic fractional flow reserve (angioFFR). A database of 27,000 synthetic coronary anatomy models was created; 90% of the models were used for training the prototype and 10% were used for verification. 3D, 3-dimensional; CFD, computational fluid dynamics; FFR, fractional flow reserve.

Figure 2.

Angiographic fractional flow reserve (angioFFR) analysis. 3D, 3-dimensional; CFD, computational fluid dynamics; FFR, fractional flow reserve.

This study aimed to assess the diagnostic accuracy of the non-commercial prototype software angioFFR in detecting hemodynamically relevant CAD using invasive FFR as the reference standard. We also assessed the FFR gradients of the pre-stenting segments derived from invasive FFR and from the angioFFR application.

Methods

Patients

This study was a prospective single-center study in Japan. Patients with 30–90% angiographic stenoses in major epicardial vessels were included.^1,2,11 Patients who underwent coronary artery bypass grafting and patients with myocardial infarction for <72 h, left main trunk disease, ostial lesions, severe aortic valve stenosis, or vessels with stents or collaterals were excluded.¹¹ Written informed consent was obtained from patients before enrollment in the study. The study was approved by the Ethics Committee of Gifu Heart Center (No. 2018002), has been registered with the University Hospital Medical Information Network (UMIN) Clinical Trials Registry (ID: UMIN000041712), and was performed in accordance with the Declaration of Helsinki.

Angiography

All procedures were performed in accordance with the guidelines using a 5- or 6-Fr system.¹² A biplane angiography system (Artis zee biplane; Siemens Healthcare GmbH) was used for fluoroscopic imaging throughout the procedure. The imaging parameters were 78 kV and 15 frames per s (fps). After the administration of intracoronary nitrates (300 μg), contrast injections were performed manually in 5-mL pulses (370 mg iodine/mL [Iopamiron; Bayer HealthCare Co, Ltd. Osaka, Japan] or 350 mg iodine/mL [Omnipaque; Daiichi Sankyo Co., Ltd., Tokyo, Japan]). The stenotic vessels were visually classified as focal, tandem, diffuse, bifurcation, severe calcification, and severe vessel tortuosity as described previously.¹³ Quantitative coronary angiography (QCA; Heart II^®; RATOC SYSTEM ENGINEERING, Tokyo, Japan) was used to assess lesion severity.¹⁴

Invasive FFR Measurement

Invasive FFR was measured after administration of intracoronary nitrates.⁵ A pressure wire sensor (PressureWire^TM X [Abbott Vascular, Santa Clara, CA, USA], OptoWire^TM [Opsens Inc., Quebec City, Canada], Verrata Plus^TM [Philips, Amsterdam, Netherlands], Comet^TM wire [Boston Scientific, Minneapolis, MN, USA], or Navvus RXi^TM [ACIST Medical Systems, Eden Prairie, MN, USA]) was located at least 20 mm distal to the stenosis.⁵ FFR was measured under maximum hyperemia induced by intravenous infusion of ATP (180 μg/kg/min), and an intracoronary infusion of 2 mg nicorandil or papaverine at the discretion of the physicians (12 mg for the left coronary artery; 8 mg for the right coronary artery [RCA]),⁵ and repeated if the drift exceeded ±0.02.¹⁵

AngioFFR Application

The angioFFR prototype software (not for diagnostic use) was based on AI using a supervised machine-learning algorithm for the computation of blood pressure along coronary arteries, trained upfront at the manufacturer’s site (Figure 1). For training, a database of more than 27,000 coronary representations was synthetically generated and FFR was calculated using a 3D CFD model (ANSYS CFD 12; Ansys, Inc., Canonsburg, PA, USA). In the clinical setting, the upfront-trained application of the prototype can estimate angioFFR values based on 2 angiographic scenes.

AngioFFR Analysis

Angiography data for all enrolled patients were anonymized and transferred to a stand-alone research workstation equipped with the non-commercial angioFFR prototype software (Monaco WS with angioFFR; Siemens Healthcare GmbH). In an offline analysis, angioFFR values were measured by 2 raters (H.O. [board-certified cardiologist with 10 years experience] and S.I. [radiological technologist with 15 years experience]) blinded to the invasive FFR values, to the patient’s clinical data, and to each other’s results.

The steps of angioFFR measurements were as follows. First, 2 angiographic scenes, ≥30° apart, of the interrogated vessel with minimal foreshortening were selected. Then, the diastolic phases with maximum contrast filling were detected automatically. In the absence of electrocardiogram results, AI-based heart phase detection was used to automatically determine suitable diastolic frames.¹⁶ Seed points were placed in the interrogated vessel proximal to the stenosis and at the position of the pressure wire sensor. Then, the 2 scenes were matched using anatomical landmarks, after which a side branch with a diameter of ≥2 mm was annotated (Figure 2). Within the angioFFR application, the lumen boundaries are automatically traced given a provided proximal and a distal endpoint. If necessary, the contours were manually corrected. Finally, the software displayed angioFFR values, a pullback curve, and the mean radius along the vessel (Figure 3).

Figure 3.

Case example. (Upper) Invasive fractional flow reserve (FFR) and angiographic fractional flow reserve (angioFFR) values at the pressure wire position. (Lower) Invasive FFR and angioFFR gradients at pre-stenting segments derived from each pullback curve.

Assessment of FFR Gradients Derived From Invasive FFR and AngioFFR Pullback

Invasive FFR pullback was performed in patients who were selected to undergo PCI to determine which lesion was physiologically significant and needed to be stented (Figure 3).^17,18 The pressure gradient by invasive FFR was defined as the difference in FFR values between the proximal and distal edges at the pre-stenting segments. In the same way, the pressure gradient by angioFFR was obtained using values from the angioFFR pullback curve.

Endpoints

The primary endpoint was the diagnostic accuracy of angioFFR using invasive FFR ≤0.80 as the reference standard.¹⁹ Secondary endpoints included the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) of angioFFR; the area under the curve (AUC) of the receiver operating characteristic (ROC) graphs between invasive FFR and angioFFR; the correlation between the FFR gradients of the pre-stenting segments derived from invasive FFR and angioFFR application; the reproducibility of angioFFR; and the gray zone of angioFFR, defined as a sensitivity lower than 90% or specificity lower than 90%.²⁰

Statistical Analysis

Categorical variables are expressed as frequencies and proportions. Continuous variables are expressed as the mean±SD or as the median and interquartile range (IQR), and were compared using the paired Student’s t-test or paired Wilcoxon signed-rank test. The cut-off value for 2-dimensional (2D) QCA was defined as 50%.⁸ For lesion-based analysis, the accuracy, sensitivity, specificity, PPV, and NPV, with 2D-QCA >50% and angioFFR ≤0.80, were calculated as proportions with 95% confidence intervals (95% CIs) using an invasive FFR value ≤0.80 as the reference standard;⁸ angioFFR was assessed in all vessels, left anterior descending artery (LAD), left circumflex artery (LCX), and RCA. McNemar’s test was used to compare the sensitivity and specificity between 2D-QCA and angioFFR. Moskowitz and Pepe’s tests were used to compare the PPV and NPV of the 2 diagnostic tests.²¹ The diagnostic performances of 2D-QCA and angioFFR were assessed using the AUC from ROC analysis and its 95% CI to predict invasive FFR values ≤0.80.^4,22 Per-vessel AUC comparisons were performed using the DeLong method. Correlations were assessed using Pearson’s correlation coefficient (95% CI). The agreement between invasive FFR and angioFFR was assessed using Bland-Altman analysis. Intrarater (H.O.) and interrater (H.O. and S.I.) reliabilities were assessed using intraclass correlation coefficients, scatter plots with regression lines, and Bland-Altman analysis. Statistical significance was set at P values two-tailed <0.05. Statistical analyses were performed using MedCalc version 19.3.1 (MedCalc Software Bvba, Ostend, Belgium), R (R Foundation for Statistical Computing, Vienna, Austria), and SAS version 9.4 (SAS Institute Inc., Cary, NC, USA).

Results

In all, 293 vessels in 232 consecutive patients with 30–90% angiographic stenoses who underwent invasive FFR were enrolled between November 2018 and February 2020. We excluded 33 vessels according to the exclusion criteria and, in the offline analysis, we excluded 3 vessels with insufficient contrast enhancement, 1 lesion overlapping with another vessel, and 3 severely tortuous vessels that precluded contour detection. Finally, we analyzed 253 vessels from 200 patients (Figure 4).

Figure 4.

Flowchart of patient selection. AngioFFR, angiographic fractional flow reserve; CAD, coronary artery disease; FFR, fractional flow reserve; ICA, invasive coronary angiography; PCI, percutaneous coronary intervention.

The baseline characteristics of the participants are presented in Table 1. The patients were 71±10 years old and included 143 (71.5%) men. Among the 253 vessels analyzed, tandem lesions, bifurcation lesions, lesions with heavy calcification, and severe tortuous vessels accounted for 69 (27.3%), 88 (34.8%), 17 (6.7%), and 23 (9.1%) of the lesions, respectively.

Table 1. Baseline Patient and Lesion Characteristics

Patient (n=200) characteristics
Age (years)	71±10
Male sex	143 (71.5)
BMI (kg/m2)	23.6±3.3
Hypertension	142 (71.0)
Dyslipidemia	110 (55.0)
Diabetes	76 (38.0)
Current smoker	42 (21.0)
Previous PCI	42 (21.0)
Previous MI	24 (12.0)
Multivessel disease	91 (45.5)
LVEF (%)	61.7±12.6
eGFR (mL/min/1.73 m2)	58.4±22.2
Stable angina pectoris	114 (57.0)
Unstable angina pectoris	21 (10.5)
Silent ischemia	65 (32.5)
Angiography
Contrast volume (mL)
Coronary angiography cases (n=147)	32.3±9.7
PCI cases (n=53)	80.0±20.4
Lesion characteristics	n=253
Target lesion
LAD	140 (55.3)
LCX	58 (22.9)
RCA	55 (21.7)
Focal lesion	146 (57.7)
Tandem lesion	69 (27.3)
Diffuse lesion	38 (15.0)
Bifurcation with side branch ≥2.0 mm	88 (34.8)
Severe calcification lesion	17 (6.7)
Severe vessel tortuosity	23 (9.1)
QCA
Diameter stenosis (%)	50.0±12.1
Diameter stenosis
30–50%	140 (55.3)
51–70%	99 (39.1)
71–90%	14 (5.5)
Reference diameter (mm)	2.9±1.8
Minimum luminal diameter (mm)	1.4±0.5
Length of lesions (mm)	20.6±10.2

Continuous variables are presented as the mean±SD, whereas categorical variables are presented as n (%). BMI, body mass index; eGFR, estimated glomerular filtration rate; LAD, left anterior descending artery; LCX, left circumflex artery; LVEF, left ventricular ejection fraction; MI, myocardial infarction; PCI, percutaneous coronary intervention; QCA, quantitative coronary angiography; RCA, right coronary artery.

Diagnostic Performance of AngioFFR

The distributions of angioFFR and invasive FFR values are shown in Supplementary Figure 1. The mean invasive FFR and angioFFR values were 0.83±0.10 and 0.83±0.10, respectively (P=0.47). Among the 253 lesions, 95 (37.5%) and 82 (32.4%) were found to have invasive FFR and angioFFR ≤0.80, respectively (P=0.23).

As indicated in Table 2, the accuracy of the angioFFR was 87.7% (95% CI 83.1–91.5), with a sensitivity of 76.8% (95% CI 68.4–85.3), a specificity of 94.3% (95% CI 90.7–97.9), a PPV of 89.0% (95%: CI 82.3–95.8), and an NPV of 87.1% (95% CI 82.1–92.2). The AUC of angioFFR was significantly higher than that of 2D-QCA (0.90 [95% CI 0.86–0.94] vs. 0.75 [95% CI 0.68–0.81], respectively; P<0.001; Table 2). Subgroup analysis showed that the accuracy and AUC of angioFFR for each major vessel were good: 85.0% (95% CI 78.0–90.5%) and 0.89 (95% CI 0.82–0.93) for the LAD; 94.8% [95% CI 85.6–98.9%] and 0.95 (95% CI 0.86–0.99) for the LCX; and 87.3% (95% CI 75.5–94.7%) and 0.90 (95% CI 0.78–0.96) for the RCA (Supplementary Table 1; Supplementary Figure 2).

Table 2. Diagnostic Performance of 2D-QCA and AngioFFR

	2D-QCA	AngioFFR	P value
Accuracy (%)	72.3 (66.4–77.8)	87.7 (83.1–91.5)	–
Sensitivity (%)	72.6 (63.7–81.6)	76.8 (68.4–85.3)	0.43
Specificity (%)	72.2 (65.2–79.1)	94.3 (90.7–97.9)	<0.001
PPV (%)	61.1 (52.1–70.1)	89.0 (82.3–95.8)	<0.001
NPV (%)	81.4 (75.0–87.9)	87.1 (82.1–92.2)	0.06
AUC	0.75 (0.68–0.81)	0.90 (0.86–0.94)	<0.001

Values in parentheses are 95% confidence intervals. An invasive fractional flow reserve of ≤0.80 indicates ischemia. 2D, 2-dimensional; angioFFR, angiographic fractional flow reserve; AUC, area under the receiver operating characteristic curve; NPV, negative predictive value; QCA, quantitative coronary angiography; PPV, positive predictive value.

There was a good correlation between angioFFR and invasive FFR (r=0.76; 95% CI 0.71–0.81; P<0.001; Figure 5). The agreement between invasive FFR and angioFFR was 0.003 (limits of agreement: −0.13, 0.14; Figure 5).

Figure 5.

(A) Correlation and (B) agreement between invasive fractional flow reserve (FFR) and angiographic fractional flow reserve (angioFFR). (C) Area under the receiver operating characteristic curve (AUC) of angioFFR for the identification of invasive FFR ≤0.80. CI, confidence interval.

The gradient of invasive FFR at the pre-stenting segments was comparable to that of angioFFR (n=51; Table 3).

Table 3. Comparison of the Gradient Derived From Invasive FFR and AngioFFR

	Invasive FFR	AngioFFR	P value
FFR gradient at pre-stenting segments	0.22±0.10	0.22±0.11	0.87

Data are given as the mean±SD. Data are for 51 of 56 segments where were underwent percutaneous coronary intervention; there was no fractional flow reserve (FFR) pullback for 5 patients. AngioFFR, angiographic fractional flow reserve.

The intra- and interrater intraclass coefficients were 0.82 (95% CI 0.66–0.90) and 0.83 (95% CI 0.68–0.91), respectively. Scatter plots with regression lines and Bland-Altman plots are shown in Supplementary Figure 3.

Gray Zone of AngioFFR

AngioFFR values ≤0.80 and ≥0.88 yielded a sensitivity of 92% and specificity of 94%. Based on these findings, we set the gray zone as 0.81–0.87. The accuracy of the angioFFR without the gray zone was 90.4% (95% CI 85.3–94.2%), with a sensitivity of 88.0% (95% CI 79.0–94.1%), a specificity of 92.4% (95% CI 85.5–96.7%), a PPV of 90.1% (95%: CI 82.4–94.7%), an NPV of 90.7% (95% CI 84.4–94.6%), and an AUC of 0.94 (95% CI 0.90–0.97%). The implementation of a gray zone would have allowed the use of pressure wire to be spared in 188 of 253 vessels (74.3%).

There were 10 (4.0%) false-negative cases beyond the gray zone (angioFFR ≥0.88 and invasive FFR ≤0.80; Supplementary Table 2). There were 8 (3.2%) false-positive cases beyond the gray zone (angioFFR ≤0.80 and invasive FFR ≥0.81; Supplementary Table 3).

Discussion

This study was a single-center prospective study assessing the diagnostic performance of AI-based angioFFR. The main findings of this study are that: (1) angioFFR had good diagnostic accuracy for detecting hemodynamically significant CAD using invasive FFR as the reference standard; (2) the FFR gradients of pre-stenting segments derived from invasive FFR and the angioFFR application were comparable; and (3) the reproducibility of angioFFR was good.

Diagnostic Accuracy of AngioFFR

AI-based angioFFR had high accuracy (87.7%) for detecting hemodynamically relevant stenoses. The diagnostic accuracy of angioFFR (r=0.76; mean difference 0.003; AUC 0.90) appeared to be similar to that of other angiography-derived FFR applications (r=0.81; mean difference −0.003; AUC 0.84).⁶ The unique feature of the angioFFR is its high specificity (94.3%; i.e., the previously reported pooled specificity was 90%).⁶ This finding means that the angioFFR can detect functionally significant lesions for revascularization with high probability. The sensitivity of the angioFFR (76.8%) was numerically lower than the previously reported pooled specificity (89%).⁶ A potential explanation for this is that the unclear imaging due to severe stenosis, calcified lesions, or ulcer lesions caused overestimation because contour detection of the vessel may become larger.

Reliability of the AngioFFR Pullback Curve

The FFR gradients in the pre-stenting segments derived from invasive FFR and the angioFFR application were comparable. In PCI, the angioFFR pullback curve is useful for the determination of physiologically significant lesions. These findings also mean that the precision of the 3D construction of coronary angiography using the angioFFR application leads to high accuracy of angioFFR values and pullback.

How to Use AngioFFR in the Clinical Setting

Without a pressure wire, a discordant case cannot be recognized in the clinical setting. Currently angioFFR indicates lesions that are likely to be negative or positive. If the stenosis falls in the gray zone of angioFFR (0.81–0.87), a pressure wire should be used. The methods using the gray zone increased accuracy (from 87.7% to 90.4%) and decreased the use of pressure wire (74.3%); hence, setting a gray zone may be effective in clinical scenarios. The gray zone of angioFFR appeared to be higher than that of invasive FFR (0.75–0.80). A previous meta-analysis of angiography-derived FFR showed that the gray zone of angiography-derived FFR was between 0.77 and 0.86, which the clinician needs to be aware of.⁶

In the present study, the false-negative cases beyond gray zone (angioFFR ≥0.88 and invasive FFR ≤0.80) were between 0.75 and 0.80 on invasive FFR (Supplementary Table 2). They were the gray zone on invasive FFR, so optimal medical therapy definitely needs to be applied even if the angioFFR was ≥0.88 because the presence of plaques cannot be ruled out. For false-positive cases beyond the gray zone (angioFFR ≤0.80 and invasive FFR ≥0.81), the values of angioFFR were between 0.75 and 0.79 (Supplementary Table 3). When angioFFR is between 0.75 and 0.80, whether the lesions should be stented needs to be carefully considered, as does the possibility that the lesions correspond to functional focal lesions derived from the angioFFR pullback curve. Further studies that compare invasive FFR- and angioFFR-guided strategies using the gray zone are warranted.

Clinical Implications

AI-based angioFFR has several advantages: (1) angioFFR is derived from a robust algorithm that was trained with 3D data (>27,000 coronary anatomies), including CFD results, which ensures that almost all lesions can be analyzed; (2) AI-based FFR computation has the benefit of speeding up computation for real-time use with the accuracy of real 3D CFD computations (in contrast, non-AI-based FFR may reduce accuracy because the CFD computation model would need to be reduced to shorten the long computation time); (3) the AI algorithm can be further refined by additional clinical data, leading to increased angioFFR accuracy; and (4) the intra- and interrater reliability for angioFFR was good; hence, angioFFR values can be reliably used by different users.

A disadvantage in the development of all AI-based applications is that the first training of the algorithms is often done on synthetic vessel geometries and requires large amounts of clinical data for validation. Hence, in vivo studies are important. The raw data will be input to further increase the accuracy of the angioFFR computation. We believe that angioFFR has the potential to become a useful tool in daily clinical practice.

Study Limitations

This study has limitations. First, this study used data from patients from a single center, which may have led to selection bias. Second, 40 of 293 (13.7%) patients met the exclusion criteria, therefore the findings of this study may not be generalizable. Third, the processing time was not measured because a prototype was used. The angioFFR application uses AI instead of CFD, therefore the processing time is expected to be short. However, further studies are needed. Finally, clinical outcome studies of angioFFR-guided PCI are warranted to assess its clinical usefulness.

Conclusions

AI-based angioFFR has good diagnostic accuracy for detecting hemodynamically relevant stenosis using invasive FFR as the reference standard. The gradients of invasive FFR and angioFFR in the pre-stenting segments were comparable.

Acknowledgments

The authors thank radiology technicians Syunsuke Imai, Yuki Nakashima, Takahito Mizusaki, and Takuya Kagatani. The authors also thank medical engineers Shinya Iwata, Airi Kimata, and Tetsuya Hiraiwa for quantitative analysis of the coronary angiography. The authors also express their heartfelt gratitude to the Japan Society of Clinical Research for their dedicated support.

Sources of Funding

This study was supported, in part, by a research grant (ID C00224265) from Siemens Healthcare K.K. Siemens Healthcare provided a workstation equipped with the non-commercial angioFFR prototype (not for diagnostic use) on loan. The concepts and prototype software presented in this study are based on research. Due to regulatory reasons, its future availability cannot be guaranteed. Siemens Healthcare was not involved in the conduct of the study, nor in data collection, analysis, or interpretation.

Disclosures

The authors declare no conflicts of interest associated with this manuscript.

IRB Information

This study was approved by the Ethics Committee of Gifu Heart Center (No. 2018002), has been registered with the University Hospital Medical Information Network (UMIN) Clinical Trials Registry (ID: UMIN000041712), and was performed in accordance with the Declaration of Helsinki.

Data Availability

The deidentified participant data will not be shared.

Supplementary Files

Please find supplementary file(s);

https://doi.org/10.1253/circj.CJ-22-0771

References

• 1.   Patel MR, Calhoon JH, Dehmer GJ, Grantham JA, Maddox TM, Maron DJ, et al. ACC/AATS/AHA/ASE/ASNC/SCAI/SCCT/STS 2017 appropriate use criteria for coronary revascularization in patients with stable ischemic heart disease: A report of the American College of Cardiology Appropriate Use Criteria Task Force, American Association for Thoracic Surgery, American Heart Association, American Society of Echocardiography, American Society of Nuclear Cardiology, Society for Cardiovascular Angiography and Interventions, Society of Cardiovascular Computed Tomography, and Society of Thoracic Surgeons. J Am Coll Cardiol 2017; 69: 2212–2241.
• 2.   Knuuti J, Wijns W, Saraste A, Capodanno D, Barbato E, Funck-Brentano C, et al. 2019 ESC guidelines for the diagnosis and management of chronic coronary syndromes. Eur Heart J 2020; 41: 407–477.
• 3.   van Nunen LX, Zimmermann FM, Tonino PA, Barbato E, Baumbach A, Engstrøm T, et al. Fractional flow reserve versus angiography for guidance of PCI in patients with multivessel coronary artery disease (FAME): 5-year follow-up of a randomised controlled trial. Lancet 2015; 386: 1853–1860.
• 4.   De Bruyne B, Pijls NH, Kalesan B, Barbato E, Tonino PAL, Piroth Z, et al. Fractional flow reserve-guided PCI versus medical therapy in stable coronary disease. N Engl J Med 2012; 367: 991–1001.
• 5.   Toth GG, Johnson NP, Jeremias A, Pellicano M, Vranckx P, Fearon WF, et al. Standardization of fractional flow reserve measurements. J Am Coll Cardiol 2016; 68: 742–753.
• 6.   Collet C, Onuma Y, Sonck J, Asano T, Vandeloo B, Kornowski R, et al. Diagnostic performance of angiography-derived fractional flow reserve: A systematic review and Bayesian meta-analysis. Eur Heart J 2018; 39: 3314–3321.
• 7.   Trobs M, Achenbach S, Rother J, Redel T, Scheuering M, Winneberger D, et al. Comparison of fractional flow reserve based on computational fluid dynamics modeling using coronary angiographic vessel morphology versus invasively measured fractional flow reserve. Am J Cardiol 2016; 117: 29–35.
• 8.   Westra J, Andersen BK, Campo G, Matsuo H, Koltowski L, Eftekhari A, et al. Diagnostic performance of in-procedure angiography-derived quantitative flow reserve compared to pressure-derived fractional flow reserve: The FAVOR II Europe–Japan Study. J Am Heart Assoc 2018; 7: e009603.
• 9.   Fearon WF, Achenbach S, Engstrom T, Assali A, Shlofmitz R, Jeremias A, et al. Accuracy of fractional flow reserve derived from coronary angiography. Circulation 2019; 139: 477–484.
• 10.   Itu L, Rapaka S, Passerini T, Georgescu B, Schwemmer C, Schoebinger M, et al. A machine-learning approach for computation of fractional flow reserve from coronary computed tomography. J Appl Physiol (1985) 2016; 121: 42–52.
• 11.   Li J, Gong Y, Wang W, Yang Q, Liu B, Lu Y, et al. Accuracy of computational pressure-fluid dynamics applied to coronary angiography to derive fractional flow reserve: FLASH FFR. Cardiovasc Res 2020; 116: 1349–1356.
• 12.   Naidu SS, Rao SV, Blankenship J, Cavendish JJ, Farah T, Moussa I, et al. Clinical expert consensus statement on best practices in the cardiac catheterization laboratory: Society for Cardiovascular Angiography and Interventions. Catheter Cardiovasc Interv 2012; 80: 456–464.
• 13.   Sianos G, Morel MA, Kappetein AP, Morice M, Colombo A, Dawkins K, et al. The SYNTAX score: An angiographic tool grading the complexity of coronary artery disease. EuroIntervention 2005; 1: 219–227.
• 14.   Pompa J, Almonacid A, Burke D. Qualitative and quantitative coronary angiography. In: Topol EJ, Teirstein PS, editors. Textbook of interventional cardiology, 6th edn. Philadelphia: Saunders, 2011; 757–775.
• 15.   Cook CM, Ahmad Y, Shun-Shin MJ, Nijjer S, Petraco R, Al-Lamee R, et al. Quantification of the effect of pressure wire drift on the diagnostic performance of fractional flow reserve, instantaneous wave-free ratio, and whole-cycle Pd/Pa. Circ Cardiovasc Interv 2016; 9: e002988.
• 16.   Ciusdel C, Turcea A, Puiu A, Itu L, Calmac L, Weiss E, et al. Deep neural networks for ECG-free cardiac phase and end-diastolic frame detection on coronary angiographies. Comput Med Imaging Graph 2020; 84: 101749.
• 17.   De Bruyne B, Pijls NH, Heyndrickx GR, Hodeige D, Kirkeeide R, Gould KL. Pressure-derived fractional flow reserve to assess serial epicardial stenoses: Theoretical basis and animal validation. Circulation 2000; 101: 1840–1847.
• 18.   Pijls NH, De Bruyne B, Bech GJ, Heyndrickx GR, Bonnier HJ, Koolen JJ. Coronary pressure measurement to assess the hemodynamic significance of serial stenoses within one coronary artery: Validation in humans. Circulation 2000; 102: 2371–2377.
• 19.   Omori H, Witberg G, Kawase Y, Tanigaki T, Okamoto S, Hirata T, et al. Angiogram based fractional flow reserve in patients with dual/triple vessel coronary artery disease. Int J Cardiol 2019; 283: 17–22.
• 20.   Cannesson M, Le Manach Y, Hofer CK, Goarin JP, Lehot JJ, Vallet B, et al. Assessing the diagnostic accuracy of pulse pressure variations for the prediction of fluid responsiveness: A “gray zone” approach. Anesthesiology 2011; 115: 231–241.
• 21.   Moskowitz CS, Pepe MS. Comparing the predictive values of diagnostic tests: Sample size and analysis for paired study designs. Clin Trials 2006; 3: 272–279.
• 22.   Tonino PA, De Bruyne B, Pijls NH, Siebert U, Ikeno F, Veer M, et al. Fractional flow reserve versus angiography for guiding percutaneous coronary intervention. N Engl J Med 2009; 360: 213–224.