Abstract
Background
Although invasive measurement of fractional flow reserve (FFR) is recommended to guide revascularization, its routine use is underutilized. Recently, a novel non-invasive software that can instantaneously produce FFR values from the diagnostic angiograms, derived completely from artificial intelligence (AI) algorithms has been developed. We aim to assess the accuracy and diagnostic performance of AI-FFR in a real-world retrospective study.
Methods
Retrospective, three-center study comparing AI-FFR values with invasive pressure wire–derived FFR obtained in patients undergoing routine diagnostic angiography. The accuracy, sensitivity, and specificity of AI-FFR were analyzed.
Results
A total of 304 vessels from 297 patients were included. Mean invasive FFR was 0.86 vs. 0.85 AI-FFR (mean difference: −0.005, P = 0.159). The diagnostic performance of AI-FFR demonstrated sensitivity of 91%, specificity 95%, positive predictive value 83% and negative predictive value 97%. Overall accuracy was 94% and the area under curve was 0.93 (95% CI 0.88–0.97). 105 lesions fell around the cutoff value (FFR = 0.75–0.85); in this sub-group, AI-FFR demonstrated sensitivity of 95%, and specificity 94%, with an AUC of 0.94 (95% CI 88.2–98.0). AI-FFR calculation time was 37.5 ± 7.4 s for each angiographic video. In 89% of cases, the software located the target lesion and in 11%, the operator manually marked the target lesion.
Conclusion
AI-FFR calculated by an AI-based, angio-derived method, demonstrated excellent diagnostic performance against invasive FFR. AI-FFR calculation was fast with high reproducibility.
Keywords: artificial intelligence, fractional flow reserve, quantitative coronary analysis
Introduction
Over the past two decades, the use of invasive intracoronary measurement of fractional flow reserve (FFR) to evaluate the physiological significance of coronary lesions and to guide revascularization strategy has been shown to improve clinical outcomes [1–3] and is recommended in both the ESC and ACC guidelines [4,5]. Despite clinical recommendations, the routine use of FFR is underutilized [6,7]. Possible reasons for FFR underuse include increased procedural time, increased hospital expenses and associated procedural risk with the use of invasive coronary wires and of injected hyperemic agents. Consequently, the search for faster and safer indices to guide revascularization has been active for more than a decade. Initially, invasive nonhyperemic physiologic metrics were developed, aimed at improving safety and ease of use [8–11]. While simpler to use, these novel techniques also remain underutilized.
In the last few years, non-invasive techniques for measuring FFR by employing computational flow dynamics (CFD) calculations from specific coronary angiograms have been developed aiming to address the limitations of invasive techniques [12–16]. These angiogram-based techniques have specific procedural and post-procedural requirements as well as a relatively long calculation time. The operator needs to acquire specific 2–3 projections of each coronary vessels while table panning is forbidden, and the software requires manual marking of a specific blood vessel segment as well as manual input of aortic pressures; only then calculation begins and typically takes ~4–6 min [17,18].
Recently, a novel non-invasive software that can instantaneously produce FFR values derived completely from artificial intelligence (AI) and machine learning-based algorithms was developed (AutocathFFR – MedHub Ltd., Tel-Aviv, Israel). AutocathFFR operates based on machine learning and deep learning approaches, utilizing Convolutional Neural Networks, allowing image and video analysis tasks. The software progresses through a multi-component algorithm allowing for vessel segmentation and lesions detection followed by AI-FFR calculations within less than a minute (Fig. 1).
Fig. 1.
Steps of AI-FFR calculation using the AutocathFFR software.
During the first steps of the analysis, the software verifies that the angiogram images are of sufficient quality and meet the software acceptance criteria for analysis. Thereafter the best images are selected by the software. This step was developed by training the algorithm with more than 13 000 invasive coronary angiography procedures comprising millions of single images. The algorithm was repeatedly trained to identify and extract distinctive ‘features’ from the training dataset. These features refer to visual patterns such as edges, shapes, textures, grayscale, and other vessel-specific characteristics, which serve as the defining attributes or distinctive elements of the images. This allowed the software to identify and extract the best frames with the maximum amount of contrast agent and to define the diastolic phase.
For the next steps, the AutocathFFR was trained in a supervised learning setup using >10 000 labeled angiography frames. These frames, where the anatomical structures, vessel names, and locations of stenoses were pre-marked by experts, assisted the AI in associating the specific visual features with the corresponding labels. This enabled vessel structure extraction followed by anatomic vessel identification and eventually stenotic lesion detection.
Finally, the AI-FFR estimator was developed using invasive FFR measurements of more than 1500 patients. The FFR estimator was built and trained using the features that are extracted during the analysis process in conjunction with the invasive FFR result. At the conclusion of training the algorithm assembled 150 different features that were extracted from each best frame during the analysis process for assessing each AI-FFR value (further explanation regarding data training process of the AI software can be found in the supplemental data, Supplemental digital content 1, http://links.lww.com/MCA/A614).
At the end of the process, the FFR estimator can regress the FFR number non-invasively based on all the information gathered along the procedure from different angiography angles and combining the information extracted from the features of each lesion from multiple views.
The AutocathFFR software can analyze the entire coronary tree in less than 1 minute with minimal requirements, making it possible to incorporate its use as part of the routine diagnostic coronary angiogram procedure. A proof-of-concept feasibility study demonstrated high accuracy rate of this AI-derived FFR compared with the classic invasive FFR [19]. In the current study, we assessed the accuracy and diagnostic performance of AI-derived FFR using the AutocathFFR software in a larger multi-center cohort.
Methods
Study design and population
This is a retrospective, three-center study comparing the accuracy of AI-derived FFR values (AI-FRR), using the AuthcathFFR software (MedHub-AI, Tel-Aviv, Israel), with invasive pressure wire–derived FFR. We analyzed accuracy, sensitivity, and specificity, where an FFR ≤ 0.80 was scored ‘positive’, and an FFR > 0.8 was considered ‘negative’.
Data of all patients undergoing FFR analysis, between October 2019 and August 2020, at three medical centers in Israel (Rambam Heath Care Campus in Haifa, Hillel Yaffe Medical Center in Hadera, and Soroka Medical Center in Beer Sheva) were collected. The clinical decision to perform FFR was at the discretion of the interventional cardiologist. The study protocol was approved by the Institutional Review Boards or Ethics Committees at each site. The clinical trial registration number is NCT04861519.
Inclusion criteria included: Subjects >18 years old with stable angina pectoris, unstable angina pectoris or non-ST-segment elevation myocardial infarction who underwent a clinically indicated invasive coronary angiogram (ICA), and in whom invasive FFR was performed to assess a non-culprit lesion in at least one coronary artery. Exclusion criteria included clinical scenarios and coronary anatomies in which invasive FFR is not fully validated (see supplemental data, Supplemental digital content 1, http://links.lww.com/MCA/A614 for full list).
Quality assurance of angiograms and FFR tracing as well as quantitative coronary angiography analysis (QCA) and acquisition of AI-FFR values with the AutocathFFR software were performed in a central independent Core Laboratory (MedStar Cardiovascular Research Network, Washington, DC, USA).
Invasive coronary angiography
All images were obtained at a cine frame rate of at least 7 frames per second. The system requires a minimum of 2 different angiographic views for each lesion of interest. However, no specific angles are mandated. The angiographic images were analyzed at the core laboratory using a QCA (Pie Medical Imaging CAAS Workstation 8.1, Maastricht, The Netherlands).
Fractional flow reserve acquisition
Invasive FFR was measured according to current guidelines [4,5]. Hyperemia was induced with either intravenous adenosine (140–180 µg/kg/min) or intracoronary adenosine (100 µg for the right coronary artery or 200 µg for the left coronary artery). The pressure recording from the pressure wire was equalized to the pressure recording from the guiding catheter with the pressure sensor positioned at the ostium of the guiding catheter. The wire was advanced to at least 20 mm beyond the stenosis being interrogated, where its position was recorded. Blood pressure tracings of the FFR acquisition were collected and reviewed by the core laboratory for assessment of the quality and ascertainment of the FFR value. The FFR systems with which the data was acquired were St. Jude/Gamida (176 vessels), Philips Volcano (89 vessels), and Opsens (39 vessels).
AI-FFR analysis
The core laboratory was trained to perform AI-FFR analysis using the AutocathFFR software in accordance with the device use instructions. The DICOM records of the diagnostic phase of the ICA were transferred directly to a dedicated computer. The software then presents on the screen each selected stenosis marking it with the acquired relevant AI-FFR. If the operator wishes to select a different specific area along the coronary tree that was not selected by the software, any lesion can be manually marked, and the software will calculate the relevant AI-FFR value.
The AI-FFR values were generated by the core laboratory study investigator, who was blinded to the invasive FFR measurement for each stenosis. An AI-FFR ≤ 0.80 was considered ‘positive’, while an AI-FFR > 0.8 was considered ‘negative’. If the software could not calculate AI-FFR, the result was considered as ‘negative’, as the software did not detect a significant lesion in the segment. The AI-FFR was compared to the ‘gold standard’, invasive FFR obtained from the wire measurement, at the same location (Fig. 2).
Fig. 2.
Example of 2 cases of AI-FFR calculation. Examples of AI-FFR calculation and pressure wire-derived FFR measurements at the same vessel. (a) Case of a negative result and B. case of a positive result.
Statistical analysis
Baseline characteristics of the study cohort are presented as mean ± SD or median with interquartile range, as appropriate. Standard summary statistics were used, suitable for the current analysis. The normality of measured variables was tested using the Kolmogorov–Smirnov test.
The AI-FFR result for each lesion was classified as true or false positive or true or false negative (using invasive FFR as the reference standard, where abnormal value was defined as ≤0.80, and normal value as >0.80 for both modalities). Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV) and accuracy of AI-FFR were calculated, and 95% CIs were added.
Agreement assessment (within subject) between AI-FFR and invasive FFR was tested by the paired sample t-test. The correlations between the two methods were evaluated using the Pearson R correlation coefficients. Agreement between the two methods was evaluated by Bland–Altman plots depicting mean differences and corresponding 95% limits of agreement. Receiver-operating curves (ROC) and measurements of the area under the ROC curve (AUC) were used to assess the performance and diagnostic ability of AI-FFR to discriminate between positive and negative invasive FFR measurements.
To measure reproducibility of the AutocathFFR software, all cases that were calculated were analyzed twice. To measure the repeatability of the AutocathFFR software, the 26 cases that needed manual identification of the lesions were analyzed three times by different operators. Data were analyzed by paired t-test, and the mean differences were point estimates with 95% CIs. The significance levels were set at 0.05. Statistical analyses were performed using the SPSS statistical package version 25 (IBM, Armonk, North Castle, NY, USA).
Results
Baseline characteristics
A total of 343 vessels of 330 patients studied at three medical centers in Israel between October 2019 and August 2020 (Rambam Health Care Campus 192 patients, Soroka Medical Center 115 patients and Hillel Yaffe Medical Center 23 patients) were included in the analysis. Twenty-six patients who did not meet the inclusion/exclusion criteria were excluded, 10 vessels did not fit the angiographic eligibility criteria and 3 cases did not fit the invasive FFR eligibility criteria. In all but one case, adenosine was administered intracoronary (intravenous administration is rarely used in Israel due to cost). A total of 304 vessels of 297 patients were included in the final analysis; the study flowchart is depicted in Fig. 3.
Fig. 3.
Study flow chart.
The cohort included 78% male patients at a mean age of 63 years, of whom 48% were people with diabetes and 44% presented with acute coronary syndrome. The LAD was the most investigated vessel, with a total number of 224 vessels (73%). Seven patients underwent evaluation of 2 separate vessels. All lesions had normal TIMI flow grade 3, 80% of lesions were calcified and 40% involved bifurcation. The QCA analysis of the lesions demonstrated mean lesion length and reference vessel diameter of 12.0 ± 6.2 and 2.7 ± 0.6 mm, respectively, with mean degree of stenosis of 47.6 ± 9.9%. The baseline characteristics of patients and vessels are reported in Table 1.
Table 1.
Baseline characteristics (patients and lesions)
| Patients (n = 297) | |
|---|---|
| Age (years) | 63.7 ± 10.0 |
| Male | 234 (78.8) |
| BMI (kg/m2) | 29.0 ± 5.0 |
| Diabetes mellitus | 127 (42.7) |
| Smoking | 90 (30.3) |
| Radial access | 281 (94.6) |
| Previous PCI | 101 (34.0) |
| Presentation | |
| Stable angina | 47 (15.8) |
| Unstable angina | 82 (27.6) |
| NSTEMI | 49 (16.5) |
| No angina | 110 (37.0) |
| Lesions (n = 304) | |
|---|---|
| Lesion length (mm) | 12.0 ± 6.2 |
| Minimum lumen diameter (mm) | 1.4 ± 0.4 |
| Reference vessel diameter (mm) | 2.7 ± 0.6 |
| Diameter stenosis (%) | 47.6 ± 9.9 |
| Vessel evaluated | |
| LAD | 224 (73.6) |
| LCX | 44 (14.6) |
| RCA | 36 (11.6) |
| Lesion location | |
| Ostial | 28 (9.2) |
| Proximal | 119 (39.1) |
| Mid | 91 (29.9) |
| Distal | 66 (21.7) |
| Bifurcation | |
| None | 183 (60.2) |
| Any bifurcation type | 121 (39.8) |
| Lesion arrangement | |
| No tandem observed | 291 (95.7) |
| Sequential | 13 (4.3) |
| Calcification | |
| None | 241 (79.3) |
| Moderate | 45 (14.8) |
| Severe | 18 (5.9) |
| Tortuosity | |
| None | 237 (78.0) |
| Moderate | 47 (15.5) |
| Severe | 20 (6.6) |
Values are presented as n (%) or mean ± SD.
Invasive FFR and AI-FFR performance
Of the 304 vessels evaluated, the mean pressure wire-derived invasive FFR value was 0.86 ± 0.08, of which 235 (77%) vessels had negative FFR (>0.80) values while 69 (23%) vessels had positive FFR (≤0.80) values. 105 (35%) vessels had FFR values around the cutoff point between 0.75–0.85.
A total of 304 vessels were analyzed by AutocathFFR software (using only the diagnostic angiograms). In 270 (88.9%) cases, the software located and calculated AI-FFR values of the target lesion. In 34 (11.1%) cases, the operator needed to mark the target lesion manually for AI-FFR calculation. In 8 cases that were manually marked, the software could not calculate the AI-FFR in the manually selected segment. These 8 cases were re-analyzed by the sponsor– in 6 of these cases reselecting the lesion produced an AI-FFR value that agreed with the invasive FFR value, and in 2 cases the algorithm could not identify an anatomic lesion, indicating non-significant disease. Indeed these 2 cases had negative invasive FFR values.
The running time of AI-FFR calculation was 37.5 ± 7.4 s for each angiographic video. No device malfunctions were reported throughout the AI-FFR analysis.
The mean AI-FFR value was 0.85 ± 0.06 vs. 0.86 ± 0.08 measured by the invasive pressure wire FFR (mean difference: −0.005, P = 0.159). The distribution of AI-FFR values vs. invasive FFR value are shown in Fig. 4a.
Fig. 4.
Agreement between AI-FFR and invasive FFR. (a) Scatterplot of AI-FFR vs. invasive FFR. (b) Bland–Altman plot, with horizontal lines representing mean ± 2 SD.
Diagnostic performance of AI-FFR compared to invasive FFR
The sensitivity and specificity of AI-FFR as compared to invasive FFR were 91.3% (95% CI 0.82–0.97) and 94.5% (95% CI 0.91–0.97), respectively. PPV was 0.83 (95% CI 0.74–0.91), while the NPV was 0.97 (95% CI 0.95–0.99) (Table 2). The ROC curve and area under the ROC curve (AUC) for the overall population showed high AI-FFR performance (AUC = 0.929) (Fig. 5).
Table 2.
Diagnostic performance of AI-FFR vs. invasive FFR
| AI-FFR performance in the overall population N = 304 (vessels) |
AI-FFR performance around the cutoff (0.75–0.85) N = 105 (vessels) |
|
|---|---|---|
| Accuracy | 93.7% | 94.3% |
| AUC | 0.93 (0.88–0.97) | 0.94 (0.88–0.98) |
| Sensitivity | 91.3% (82.0–97.0) | 95.1% (83.0–99.5%) |
| Specificity | 94.5% (91.0–97.0) | 93.8% (84.6–98.0%) |
| Positive predictive value | 82.9 (72.5–90.6) | 90.7 (77.8–96.9) |
| Negative predictive value | 97.4 (94.4–99.0) | 96.8 (88.3–99.8) |
Data between brackets is 95% confidence intervals.
AUC, area under the curve.
Fig. 5.
Receiver operating characteristic (ROC) curve analysis of AI-FFR vs. Invasive FFR measurements.
Comparison of the AI-FFR vs. invasive pressure wire-based FFR measurements found a goodness of fit of R2 = 0.281 (Fig. 4a). Regression analysis of the two methods measured a regression slope of 0.689 (P < 0.001) and intra-class correlation coefficient of 0.512 (95% CI: 0.569–0.743). Paired sample statistics found no significant difference between the two methods (P < 0.16). Similarly, the Bland–Altman plot (Fig. 4b) demonstrated high agreement between AI-FFR and invasive FFR [mean of 0.005 (SD 0.066) with upper and lower limits of agreement 0.136 and −0.123, respectively].
When analyzing AI-FFR results per gender sub-group, the diagnostic performance was better in females (mean AI-FFR 0.86 vs. invasive FFR 0.88 (P = ns); sensitivity 100%, specificity 96.4%, accuracy 97.0%, PPV 84.6%, and NPV 100%) than in males (mean AI-FFR 0.85 vs. invasive FFR 0.85 (P = ns); 89.7%, 93.9%, 92.9%, 82.5%, 96.6%, respectively). AI-FFR performance by vessel type and segment can be found in the supplemental data, Supplemental digital content 1, http://links.lww.com/MCA/A614.
Performance of AI-FFR in values around the cutoff value
There were 105 lesions that fell around the cutoff value (FFR = 0.75–0.85); in these, the mean AI-FFR was 0.83 and the mean invasive FFR was 0.81. The sensitivity of AI-FFR for results around the cutoff value was 95.1% (95% CI 83.5–99.4%), the specificity was 93.8% (95% CI 84.8–98.3%), and the AUC was 0.944 (P < 0.0001) and 95% CI of 0.882–0.980 (Table 2).
Reproducibility and repeatability
Reproducibility analysis was performed using the data of the 270 in which the AutocathFFR software has identified the target lesions and calculated AI-FFR. These 270 cases were then re-analyzed by the AutocathFFR software. The mean FFR value was the same for the two software runs (0.85). Hence full agreement (100% reproducibility) was achieved in all measurement pairs. Repeatability analysis was performed for the 26 vessels in which AI-FFR was obtained manually. These cases were re-analyzed by 3 different operators. The overall AI-FFR means for readers 1, 2, and 3 were: 0.85 ± 0.06, 0.85 ± 0.05, and 0.85 ± 0.05, P = 0.436.
Discussion
In this retrospective, multi-center study, the performance of a novel non-invasive and fully automated software that can calculate FFR values from diagnostic angiograms based on AI analysis was assessed against pressure wire-derived FFR. The main findings of our study are: The AI-FFR showed high diagnostic accuracy compared to invasive FFR (sensitivity of 91.3% and specificity of 94.5%), with excellent reproducibility and repeatability. The overall mean difference between AI-FFR and FFR was negligible and achieved a very high AUC (0.93). Importantly, the same results were observed in the patients with FFR values falling around the FFR cut point (0.75–0.85), which are the most challenging cases to assess. Females had better performance than men, but given that men comprised 78% of the study population, this may be due to the play of chance.
To date, this is the largest, core laboratory conducted, analysis of the AutocathFFR software. A previous early feasibility study [19] analyzed 31 patients and demonstrated promising accuracy of AI-FFR predicting pressure wire FFR (sensitivity, specificity, and diagnostic accuracy of 88%, 93% and 90%, respectively). The current study has not only demonstrated high accuracy, but also showed excellent speed. Images were analyzed in less than 40 s and 89% of lesions were located without need for any manual annotation.
When comparing AutocathFFR’s diagnostic performance to other currently available non-invasive angio-derived FFR softwares (such as quantitative flow ratio (QFR), vessel FFR (vFFR), and FFRangio), the AI-FFR performance was similar or better. A meta-analysis of QFR, vFFR, and FFRangio resulted in sensitivities of 84%, 87%, and 91%, and specificities of 89%, 91%, and 94%, respectively [15,16]. AI-FFR proved to have both a higher sensitivity (91.3%) and specificity (94.5%) than all three alternate methods, as well as the highest diagnostic accuracy rate at 94%, while the AUCs are comparable. The diagnostic performance of AI-FFR around the cutoff value was superior to other technologies which failed to show high sensitivities and specificities within this range of FFR [20].
There are significant differences between AutocathFFR software and other non-invasive angio-derived FFR tools. While the other tools are based on fluid dynamics algorithms or computational fluid dynamic (CFD) calculation AutocathFFR is based fully on machine learning and AI-derived algorithms. This fundamental difference directly affects important procedural and post-procedural aspects such as intra-procedural angiographic requirements, calculation time, and possibility for post-procedure offline calculation. Indeed, the CFD-derived FFR requires specific 2–3 angiographic angles during which table panning is forbidden. Thereafter manual insertion of blood pressure as well as annotation of the target lesion and the surrounding vessel contour needs to be marked by catheterization laboratory (cath-lab) personnel, and only then CFD calculation will start. Total time for annotation and calculation takes between 3 and 5 min [17], for example, in the FAVOR III China study, the QFR calculation took an average 3.9 ± 1.4 min [21]. This complex process compromises user experience, wastes valuable Cath-lab time, increases training time and learning curve and may result in suboptimal reproducibility.
In contrast to the other technologies, our data demonstrated that AI-FFR calculation is highly accurate, fast and with minimal manual inputs required from the interventional cardiologist or cath-lab personnel. As such, AI-FFR seamlessly accommodates the cath-lab routines without the need to change personal or institutional habits, and without lengthening the procedure. Moreover, the lack of specific angiographic or procedural requirements means that offline AI-FFR analysis from any previous coronary angiogram is feasible. In this way the clinician can gain physiologic data from previously performed angiograms (that may also be from an outside medical center), and better prepare for the procedure while tailoring an improved treatment plan.
Limitations
Our study is a retrospective study with known potential biases and confounders. Due to the retrospective character of the study, the analysis was performed by a Core-lab technician and not by the Cath-lab personnel. The mean invasive FFR value was 0.86 and 77% of measurements were negative (>0.80). Although these findings represent a milder diseased population compared to prospective physiological studies, the excellent performance of the software in values around the cutoff value between 0.75 and 0.85 is reassuring. Indeed, our data represents real-life practice, where the decision to perform FFR was at the discretion of the interventional cardiologist only. Our core lab could not calculate an AI-FFR value for 8 vessels. These vessels were successfully re-analyzed by the sponsor and the derived AI-FFR values were in full agreement with invasive FFR measurements; nevertheless, for the current analysis, we used the core-lab evaluation and consider the 8 vessels as negative cases. Some of the studies regarding other non-invasive angio-derived FFR tools were prospective and some included larger cohort, this should be taken into account when comparing data with our retrospective collected data. The sensitivity around the gray zone was higher than non-gray zone cases – this finding may be related to the small number of non-gray zone positive cases, introducing greater uncertainty. Finally, we followed the exclusion criteria of AngioFFR studies which excluded several angiographic scenarios in which FFR is not validated. Therefore 26 patients and 10 lesions were excluded.
Conclusion
In this multi-center retrospective study, we have shown that AI-FFR calculated by an AI-based, angio-derived FFR method, demonstrated excellent diagnostic performance against invasive FFR. Furthermore, AI-FFR calculation was fast and showed high reproducibility with minimal requirement from the interventional cardiologist allowing for potential easy integration into catheterization laboratories.
Acknowledgements
The study was funded by MedHub-AI.
Conflicts of interest
Eyal Ben-Assa, Yair Feld, Ariel Roguin, Rafael Beyar, Ori Ben-Yehuda – consultants, MedHub-AI. Justin Davies – Serves as the company president and advisor from March 2023. He has no influence on the collected data and study results. Hector M. Garcia-Garcia – No personal disclosures. Institutional grant support from Biotronik, Boston Scientific, Medtronic, Abbott, Neovasc, Shockwave, Philips, and CorFlow. For the remaining authors, there are no conflicts of interest.
Supplementary Material
Footnotes
Supplemental Digital Content is available for this article. Direct URL citations appear in the printed text and are provided in the HTML and PDF versions of this article on the journal's website, www.coronary-artery.com.
References
- 1.De Bruyne B, Pijls NHJ, 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. [DOI] [PubMed] [Google Scholar]
- 2.Pijls NHJ, van Schaardenburgh P, Manoharan G, Boersma E, Bech JW, van’t Veer M, et al. Percutaneous coronary intervention of functionally nonsignificant stenosis 5-year follow-up of the DEFER study. J Am Coll Cardiol 2007; 49:2105–2111. [DOI] [PubMed] [Google Scholar]
- 3.Patricio L, Tonino PAL, De Bruyne B, Pijls NHJ, Siebert U, Ikeno F, et al. Fractional flow reserve versus angiography for guiding percutaneous coronary intervention. N Engl J Med 2009; 360:229–230. https://pubmed.ncbi.nlm.nih.gov/19144937/. [Accessed 15 April 2023]. [DOI] [PubMed] [Google Scholar]
- 4.Neumann FJ, Sousa-Uva M, Ahlsson A, Alfonso F, Banning AP, Benedetto U, et al.; ESC Scientific Document Group. 2018 ESC/EACTS Guidelines on myocardial revascularization. Eur Heart J 2019; 40:87–165. https://pubmed.ncbi.nlm.nih.gov/30165437/. [Accessed 15 April 2023].30165437 [Google Scholar]
- 5.Lawton JS, Tamis-Holland JE, Bangalore S, Bates ER, Beckie TM, Bischoff JM, et al.; Writing Committee Members. 2021 ACC/AHA/SCAI guideline for coronary artery revascularization: executive summary: a report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol 2022; 79:197–215. https://pubmed.ncbi.nlm.nih.gov/34895951/. [Accessed 15 April 2023]. [DOI] [PubMed] [Google Scholar]
- 6.Toth GG, Toth B, Johnson NP, De Vroey F, Di Serafino L, Pyxaras S, et al. Revascularization decisions in patients with stable angina and intermediate lesions: results of the international survey on interventional strategy. Circ Cardiovasc Interv 2014; 7:751–759. https://pubmed.ncbi.nlm.nih.gov/25336468/. [Accessed 15 April 2023]. [DOI] [PubMed] [Google Scholar]
- 7.Parikh RV, Liu G, Plomondon ME, Sehested TSG, Hlatky MA, Waldo SW, et al. Utilization and outcomes of measuring fractional flow reserve in patients with stable ischemic heart disease. J Am Coll Cardiol 2020; 75:409–419. [DOI] [PubMed] [Google Scholar]
- 8.Götberg M, Christiansen EH, Gudmundsdottir IJ, Sandhall L, Danielewicz M, Jakobsen L, et al.; iFR-SWEDEHEART Investigators. Instantaneous wave-free ratio versus fractional flow reserve to guide PCI. N Engl J Med 2017; 376:1813–1823. https://www.nejm.org/doi/full/10.1056/NEJMoa1616540. [Accessed 15 April 2023]. [DOI] [PubMed] [Google Scholar]
- 9.Davies JE, Sen S, Dehbi HM, Al-Lamee R, Petraco R, Nijjer SS, et al. Use of the Instantaneous wave-free ratio or fractional flow reserve in PCI. N Engl J Med 2017; 376:1824–1834. https://www.nejm.org/doi/full/10.1056/nejmoa1700445. [Accessed 15 April 2023]. [DOI] [PubMed] [Google Scholar]
- 10.Svanerud J, Ahn JM, Jeremias A, Van ’T Veer M, Gore A, Maehara A, et al. Validation of a novel non-hyperaemic index of coronary artery stenosis severity: the Resting Full-cycle Ratio (VALIDATE RFR) study. EuroIntervention 2018; 14:806–814. https://pubmed.ncbi.nlm.nih.gov/29790478/. [Accessed 15 April 2023]. [DOI] [PubMed] [Google Scholar]
- 11.Johnson NP, Li W, Chen X, Hennigan B, Watkins S, Berry C, et al. Diastolic pressure ratio: new approach and validation vs the instantaneous wave-free ratio. Eur Heart J 2019; 40:2585–2594. https://pubmed.ncbi.nlm.nih.gov/31329863/. [Accessed 15 April 2023]. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Xu B, Tu S, Qiao S, Qu X, Chen Y, Yang J, et al. Diagnostic accuracy of angiography-based quantitative flow ratio measurements for online assessment of coronary stenosis. J Am Coll Cardiol 2017; 70:3077–3087. [DOI] [PubMed] [Google Scholar]
- 13.Fearon WF, Achenbach S, Engstrom T, Assali A, Shlofmitz R, Jeremias A, et al.; FAST-FFR Study Investigators. Accuracy of fractional flow reserve derived from coronary angiography. Circulation 2019; 139:477–484. https://www.ahajournals.org/doi/abs/10.1161/CIRCULATIONAHA.118.037350. [Accessed 8 April 2023]. [DOI] [PubMed] [Google Scholar]
- 14.Masdjedi K, Tanaka N, Van Belle E, Porouchani S, Linke A, Woitek FJ, et al. Vessel fractional flow reserve (vFFR) for the assessment of stenosis severity: the FAST II study. EuroIntervention 2022; 17:1498–1505. https://pubmed.ncbi.nlm.nih.gov/34647890/. [Accessed 15 April 2023]. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Cortés C, Carrasco-Moraleja M, Aparisi A, Rodriguez-Gabella T, Campo A, Gutiérrez H, et al. Quantitative flow ratio-meta-analysis and systematic review. Catheter Cardiovasc Interv 2021; 97:807–814. https://pubmed.ncbi.nlm.nih.gov/32196932/. [Accessed 3 January 2023]. [DOI] [PubMed] [Google Scholar]
- 16.Witberg G, De Bruyne B, Fearon WF, Achenbach S, Engstrom T, Matsuo H, et al. Diagnostic performance of angiogram-derived fractional flow reserve: a pooled analysis of 5 prospective cohort studies. JACC Cardiovasc Interv 2020; 13:488–497. [DOI] [PubMed] [Google Scholar]
- 17.Scoccia A, Tomaniak M, Neleman T, Groenland FTW, Plantes ACZ, Daemen J. Angiography-based fractional flow reserve: state of the art. Curr Cardiol Rep 2022; 24:667–678. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Morris PD, Curzen N, Gunn JP. Angiography-derived fractional flow reserve: more or less physiology? J Am Heart Assoc 2020; 9:e015586. doi: 10.1161/JAHA.119.015586. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Roguin A, Abu Dogosh A, Feld Y, Konigstein M, Lerman A, Koifman E. Early feasibility of automated artificial intelligence angiography based fractional flow reserve estimation. Am J Cardiol 2021; 139:8–14. https://pubmed.ncbi.nlm.nih.gov/33058806/. [Accessed 3 January 2023]. [DOI] [PubMed] [Google Scholar]
- 20.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. doi: 10.1161/JAHA.118.009603. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Xu B, Tu S, Song L, Jin Z, Yu B, Fu G, et al.; FAVOR III China study group. Angiographic quantitative flow ratio-guided coronary intervention (FAVOR III China): a multicentre, randomised, sham-controlled trial. Lancet 2021; 398:2149–2159. https://pubmed.ncbi.nlm.nih.gov/34742368/. [Accessed 3 January 2023]. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.