PART II - COMPUTATIONAL NON-INVASIVE PHYSIOLOGICAL ASSESSMENT OF CORONARY DISEASE
TABLE OF CONTENTS
BOOKSHOP
Updated on August 27, 2021
PART II

Computational non-invasive physiological assessment of coronary disease

Julien Adjedj1, Jelmer Westra2, Daixin Ding3, 4, Junqing Yang5, William Wijns4, Shengxian Tu3
1 Department of Cardiology, Arnault Tzanck Institute, Saint Laurent du Var, France
2 Department of Cardiology, Aarhus University Hospital, Skejby, Denmark
3 School of Biomedical Engineering, Shanghai Jiao Tong University, Shanghai, China
4 The Lambe Institute for Translational Medicine and Curam, National University of Ireland Galway, Ireland
5 Department of Cardiology, Guangdong Provincial People’s Hospital, Guangzhou, China

* The first three authors contributed equally.

Summary

Coronary computed tomography angiography (CCTA) and invasive coronary angiography (ICA) are key components in the documentation of coronary artery disease. It is however evident that sole anatomical evaluation with CCTA and ICA fails to expose the hemodynamic importance of epicardial coronary artery disease. As consequence, invasive wire-based solutions for physiological assessment (FFR, iwFR) were developed to identify ischemia-producing epicardial coronary stenosis. Despite the overall trend towards more physiology-guided revascularization, the clinical adoption of functional stenosis evaluation is inadequate. Coronary image-based computational physiology was therefore introduced for evaluation of epicardial coronary stenosis. Based on the current evidence, computational physiological assessment of coronary artery disease provides solid estimates of FFR, with the potential to improve the gatekeeper function to the catheterization laboratory, plan revascularization of epicardial coronary artery disease, optimize PCI results, and expand the use of physiology-guided coronary interventions on a global level.

Introduction

Over the last two decades, invasive pressure-derived fractional flow reserve (FFR) has become the standard test for revealing the ischemic potential of a coronary stenosis [1, 2, 3, 4, 5]. Several clinical trials have documented that FFR-guided percutaneous coronary intervention (PCI) is superior to angiography-guided PCI in terms of cost savings and clinical outcomes [6, 7]. European guidelines emphasize the importance of performing FFR or instantaneous wave-free ratio (iwFR) measurements before revascularization when prior testing for ischemia is either inconclusive or unavailable [8]. FFR measurements are objective, simple and clinically accurate, but are associated with additional time and cost, patient discomfort caused by the administration of a hyperaemic agent and a small risk of vessel injury [9]. Despite the scientific evidence and guideline-recommendations [10, 11, 12, 13] the adoption of FFR has been slow in many areas [9]. Computational non-invasive approaches have emerged as an attractive alternative to invasive FFR providing physiological assessment without dedicated intracoronary pressure wires and hyperaemic agents [14, 15, 16]. This chapter highlights the principles and practical aspects of computational non-invasive physiology.

Methodology and validation of available computational methods

Fractional flow reserve derived from coronary computed tomography angiography

Coronary computed tomography angiography (CCTA) is increasingly used as first-line diagnostic test for patients with low to intermediate pre-test risk of coronary artery disease (CAD). CCTA is cheap, safe and acts as an excellent tool to rule-out obstructive CAD. Nevertheless, CCTA alone has a poor positive predictive ability to identify flow limiting epicardial disease potentially leading to unnecessary invasive coronary angiography (ICA) [17]. As consequence, fractional flow reserve derived from CCTA (FFRCT) was developed using 3-dimensional (3D) reconstruction of coronary tree and computational fluid dynamics (CFD) [18, 19, 20]. Additionally, reduced-order CFD [21, 22, 23], machine learning [24, 25, 26], and fluid-dynamic equation-based methods [27, 28] were developed to enable on-site computation of FFR without using a super computer.

FFRCT

The computation of FFRCT is based on three main physiological assumptions: 1) the baseline coronary blood flow is proportional to the volume of myocardial mass at resting condition; 2) coronary microvascular resistance at rest is inversely associated with the subtended vessel size (morphometric law); 3) a predicable reduction of microcirculatory resistance is seen as response to infusion of vasodilators such as adenosine.

Three-dimensional geometric models of coronary tree can be obtained from CCTA images acquired from at least 64-detector row CT. Lumen boundary surface of the aorta and coronary arteries are delineated and reconstructed in 3D. Baseline coronary blood flow is estimated and applied at the inlet as boundary condition. Subsequently, hyperaemic conditions are simulated and the Navier-Stokes equations are solved using CFD, assuming blood is Newtonian fluid and the vessel wall is rigid.

Three major studies provided initial insights on the feasibility and diagnostic performance of FFRCT (DeFACTO, DISCOVER-FLOW and NXT) [18, 19, 20] ( Table 1). FFRCT is superior to conventional CCTA imaging for identification of flow limiting lesions driven by improvement of the positive predictive value. In the PACIFIC study enrolling more than 200 patients, FFRCT showed higher diagnostic performance than CCTA, SPECT (single photo emission computed tomography) and PET (positron emission tomography) to detect ischemia, using FFR as the reference standard [29] ( Table 1). FFRCT is highly reproducible with standard deviation for repeated measurement of 0.034, compared with 0.033 for FFR in a study with direct comparison of the modalities [30].

Ninety days follow up of 5000 patients in the ADVANCE registry documented a better determination of down-stream revascularization versus medical treatment for FFRCT when compared with CCTA as stand-alone modality [31]. The one year results of the ADVANCE registry illustrated that patients with FFRCT ≤0.80 had a higher event rate than patients with high FFRCT values using a restricted endpoint of cardiovascular death and myocardial infarction [32]. Follow up of patients in the NXT trial (median follow-up of 4.7 years) showed an independent association of FFRCT (0.05 interval) with major adverse cardiac events (MACE) (HR 1.4; 95% CI: 1.1, 1.8; P < .001) [33]. However, more outcome data is needed comparing FFRCT to standard practice. Based on PLATFORM and PROMISE, anatomical testing with FFRCT on top of CCTA may be a more cost-effective solution as compared to usual care and functional testing as an consequence of targeted coronary revascularization [34].

cFFR and cFFRML

The cFFR (CT-based FFR) method computes FFR from CCTA images by coupling pressure drop models to a simplified one-dimensional CFD model. A flow dynamics model that characterises the pressure drop at the stenotic segment by several stenotic features is coupled with a 1-dimensional CFD model at the non-stenotic segment to compute cFFR. The total analysis time for cFFR is between 30 to 120 minutes per case [23]. The diagnostic accuracy of cFFR approaches 75% with FFR ≤0.80 as reference ( Table 1). Image artefacts and high calcium score decreased the specificity of cFFR [23].

The recently presented cFFRML method based on machine learning algorithms automatically extracts the coronary arterial lumen and computes cFFRML [24]. The cFFRML machine-learning algorithms were derived from a synthetically generated dataset of 12000 coronary trees covering a wide variety of stenotic characteristics to describe the relationship between anatomical features and physiological parameters. The cFFRML approach demonstrates that geometric features alone can predict FFR without explicitly solving complex hemodynamic equations. The average analysis time for cFFRML was 12.4 ± 3.4 min per case in a pilot study with 2.4 ± 0.44 s attributed to sole computation time.

Machine-learning approaches are fast compared to CFD-solutions but the diagnostic performance appears to be similar ( Table 1) [24, 25, 26]. A study investigating 71 low-risk patients with CAD (mean FFR value: 0.86 ± 0.07) demonstrated that cFFRML had a diagnostic accuracy, sensitivity and specificity of 93%, 91% and 96% with FFR as reference standard [26]. However, over half of the enrolled population (77 out of 148) were excluded because of aorta-ostial lesions, bifurcation lesions, instent restenosis and tandem lesions [26].

Machine-learning approaches are highly dependent on the quality of training data. The cFFRML training data did not include the impact of atherosclerotic plaques and abrupt change in lumen borders. Further, cFFRML may not be accurate for infrequent lesion subgroups including anomalous coronary arteries and coronary aneurysms. Finally, as the machine-learning model is trained using CFD-based computational results, the diagnostic accuracy of cFFRML directly depends on the accuracy of the CFD model itself.

CT-QFR

The quantitative flow ratio (QFR) algorithm initially developed for ICA was recently adapted and applied to CCTA images for computation of FFR (CT-QFR)[27] ( Figure 1 & Table 1). All coronary arterial segments with lumen diameter over 1.5 mm by visual estimation are analysed and merged into a hierarchical tree structure. Subsequently, the reference healthy lumen as if there was no stenosis is reconstructed. Patient-specific resting coronary flow is estimated using the reference arterial volume following the allometric scaling law and converted to virtual hyperaemic flow [35]. Finally, the CT-QFR value along the coronary tree is computed based on the QFR algorithm, using the estimated hyperaemic flow as the hemodynamic boundary condition.

CT-QFR was validated in a core laboratory-based pilot study with promising results [27]. In a total of 156 vessels from 134 patients, the bias between CT-QFR and FFR was minimal (mean difference = 0.00 ± 0.06) with diagnostic concordance of 87.2% ( Table 1). The diagnostic performance of CT-QFR was at least as good as myocardial perfusion imaging in a selected second-line diagnostic strategy in patients with obstructive disease (>50% DS in ≥1 vessel) on CCTA following a post-hoc analysis of the Danish Study of Non-Invasive Diagnostic Testing in Coronary Artery Disease [28]. Total analysis time was reported as 17.1 ± 7.6 minutes for delineation of lumen boundaries for the entire coronary tree and 18.9 ± 8.5 seconds for subsequent CT-QFR computation [27]. The fact that CT-QFR can be computed without the requirement for high-performance computers indicates it as a promising tool for routinely functional assessment of coronary stenosis during CTA imaging.

Fractional flow reserve derived from invasive coronary angiography

ICA provides higher resolution compared to CCTA but still poses limited ability to guide clinical decision making when compared to FFR-guided PCI [2, 3, 4, 36, 37]. Discordance between FFR and quantitative coronary angiography (QCA) in assessment of intermediate coronary stenosis occurs in roughly one third of paired observations [36]. As consequence, the revascularization strategy changes in a substantial number of patients investigated with FFR compared to ICA alone [38] ( Table 1).

ICA-derived FFR combines functional evaluation with morphological information derived from ICA. Table 1 summarizes diagnostic performance of ICA-derived methods for key studies. A recent meta-analysis on 1842 vessels with paired ICA-derived FFR and wire-based FFR found an overall sensitivity of 89% (83%-94%) and specificity of 90% (88%-92%) with no difference between various ICA-based computational approaches [39]. However, the level of evidence differs substantially among different computational approaches and head-to-head comparisons are lacking.

QFR

Tu et al. developed quantitative flow ratio (QFR) on the basis of 3D QCA and fluid dynamic equations allowing for fast computation of FFR from conventional coronary angiography without induced hyperaemia [35] ( Figure 1 & Table 1). The main vessel is reconstructed from two angiographic projections at least 25 degrees apart. The computational model calculates the pressure drop caused by morphological variations in lumen dimensions and the estimated hyperaemic flow. The pressure at each position is then derived by subtracting the proximal pressure from the accumulated pressure drop and QFR is calculated as the distal pressure divided by the proximal pressure.

Three different blood flow models that differ in the derivation of hyperaemic flow velocity are available. The first is fixed-flow QFR (fQFR) that uses a patient-averaged hyperaemic flow velocity of 0.35m/s. The second is contrast-flow QFR (cQFR) where contrast flow velocity is derived from conventional angiography using a modified TIMI frame count. Hyperaemic flow velocity is then modelled from contrast flow based on a predicted hyperaemic response. The last model is adenosine-flow QFR (aQFR), which measures hyperaemic flow velocity based on TIMI frame counting from coronary angiography acquired during hyperaemia. The FAVOR Pilot study showed that cQFR and aQFR had better agreement with FFR compared with fQFR, and that no significant difference existed between the diagnostic accuracy of cQFR and aQFR. Thus, it is recommended to use cQFR for physiological evaluation of coronary stenosis in clinical practice because cQFR is obtained without hyperaemia.

QFR is, together with FFRangio, the only index that has been evaluated for assessment of in-procedure feasibility and diagnostic performance. In the FAVOR II study series, in-procedure QFR was computed for 96% (FAVOR II Europe-Japan) and 99% (FAVOR II China) of the FFR-interrogated vessels [40, 41]. On contrary, retrospective core-lab analysis reported feasibility rates as low as 70% mainly attributed to impaired angiographic quality [42, 43]. The latter underscores the importance of optimal image acquisition for successful derivation of computational physiology from ICA, which can be ensured by following strict acquisitions protocols [41]. The averaged total analysis time for QFR is less than five minutes, which is shorter than total time required for FFR measurement in direct comparison analysis [41]. Inter- and intra-observer variation approaches a standard deviation between 0.03 and 0.06, which is slightly higher than FFR’s of 0.02 [40, 44, 45, 46].

Recent meta-analysis including sixteen studies showed a good correlation and agreement between QFR and FFR [47]. Sensitivity, specificity, positive predictive value, and negative predictive value were 0.84, 0.89, 0.80, and 0.92, respectively [47].The diagnostic concordance between QFR and FFR is reduced in patients with microvascular dysfunction (index of microvascular resistance [IMR]>23) [48] and vessels subtending territories with prior myocardial infarction [49]. A prospective meta-analysis indicated that diabetes mellitus and severe stenosis (low FFR value or high percent diameter stenosis) are independent predictors of increased discrepancy between QFR and FFR [50].

QFR in non-culprit vessel during an acute coronary syndrome showed potential interest in terms of the evaluation of the functional significance after primary PCI. Spitaleri et al. evaluated the evolution between index and staged procedure in this context and showed good correlation and agreement between both procedures in 31 patients [51]. The largest series evaluating the efficacy of QFR in non-culprit lesions is a substudy of iSTEMI trial. In a total of 112 patients (142 lesions) with multivessel diseases, the value of QFR at index procedure was found with good correlation and agreement with QFR and FFR in staged procedure [43].

Limited data are available on the correlation between clinical outcome and ICA-derived FFR approaches, with all existing evidence from studies specifically characterizing the QFR solution [14, 51, 52, 53]. In 549 patients with stable CAD, low 3-vessel QFR (3vQFR) value was an independent predictor for MACE (HR: 0.971) [53]. A proof-of-concept study demonstrated that the QFR-derived Functional Syntax Score (fSSQFR) had higher predictive value for 2-year cardiac events compared to anatomical Syntax Score [51]. Post-hoc analysis of the randomized COMFORTABLE AMI trial showed that, in 946 non-culprit vessels left to medical therapy that were analysable by QFR, risk of 5-year clinical events was significantly higher in patients with QFR ≤0.80 versus QFR >0.80 [54]. The use of QFR in assessment of outcomes after stent implantation has been proposed in patients with stable CAD with a cut-off point of 0.89 after PCI to predict vessel oriented composite endpoint [55]. In Multicenter QFR Registry, discriminative ability of QFR for vessel-oriented composite outcomes including cardiac death, target-vessel myocardial infarction, and ischemia-driven target lesion revascularization was similar with that of FFR (0.672 versus 0.643, p = 0.147) [52].

μQFR

More recently, a new method for calculating QFR from a single angiographic view based on Murray’s bifurcation fractal law was developed [56] ( Figure 1 & Table 1). The computation of this Murray-law based QFR (μQFR) consists of several key steps [56]. Firstly, the interrogated vessel during contrast injection is delineated and the contrast flow velocity is derived based on vessel centreline length divided by the contrast filling time, followed by conversion into hyperaemic flow velocity. Subsequently, based on the frame with sharp lumen contour of the stenotic segment, the lumen boundary of both the interrogated vessel and major side branches with lumen diameter ≥1.0 mm is delineated automatically. Reference vessel diameter is reconstructed considering the step-down phenomenon across bifurcations based on Murray fractal law. Finally, pressure drop is calculated based on empirical fluid dynamic equations with the above-mentioned hyperaemic flow as the boundary condition, and μQFR is available for both the interrogated vessel and its side branches.

μQFR was validated in the FAVOR II China dataset [56]. Diagnostic accuracy, sensitivity, and specificity were 93%, 88%, and 96%, respectively. The diagnostic concordance between QCA-derived DS% and FFR was also improved based on the reconstruction of the step-down reference vessel. The overall μQFR analysis can be completed with an average time of 67 ± 22 seconds with excellent intra- (0.00 ± 0.03) and inter-observer reproducibility (0.00 ± 0.03). The use of only a single angiographic view for computation of μQFR improves the feasibility of computational FFR and is anticipated to facilitate the adoption of physiological assessment in the catheterization laboratory. Future study is warranted to compare the feasibility and diagnostic performance of single view versus two views QFR.

FFRangio

FFRangio is an ICA-derived FFR modality based on 3D vessel reconstruction and rapid flow analysis [57, 58] ( Table 1)). From three or more angiographic views, the left or right coronary tree is reconstructed in 3D. The reconstructed coronary arterial network and downstream microcirculation is modelled as an analogue circuit with each segment acting as a resistor. The resistance to flow of each segment is calculated based on arterial length and diameter. Baseline coronary blood flow is estimated based on the total volume and length of the reconstructed vasculature and hyperaemic flow is converted from baseline flow with reduced microcirculatory resistance. FFR is calculated as the ratio of the hyperaemic flow rate in the stenotic artery divided by the hyperaemic flow rate in the virtually healthy artery as if there was no stenosis.

A validation study comparing FFRangio to FFR showed good accuracy of 93% (sensitivity 88% and specificity 95%) in 184 patients (203 vessels) [57]. The FAST-FFR study validated FFRangio as measured in-procedure in a total of 319 vessels of 301 patients. FFRangio demonstrated a high in-procedure device success rate of 99% while the overall feasibility rate was 95% [58]. The diagnostic accuracy of FFRangio reached the pre-specified standard of 92% ‎in the overall population and remained high in intermediate lesions identified as FFR 0.75-0.85 (87%). ‎The per-vessel sensitivity and ‎specificity of FFRangio in predicting FFR0.80 was 94% (95% CI 88-97%) and 91% (86-95%) ( Table 1). The average computational time for FFRangio was 2.7min. However, the entire analysis time including manual adjustment was not assessed.

Vessel FFR

Masdjedi et al. applied 3D QCA and fluid dynamics equations to compute FFR (vessel FFR, vFFR) [59] ( Table 1). Three-dimensional vessel geometry is reconstructed using two angiographic image runs with 30 angle difference. The maximal hyperaemic flow is derived empirically from clinical data. The proximal coronary blood flow velocity is estimated based on the resting aortic pressure and reconstructed 3D vessel geometry with the assumption that it is preserved along the interrogated vessel segments. The trans-lesional pressure drop is calculated from viscous resistance and post-stenosis flow separation caused by hyperaemic flow.

A retrospective single-center study investigating 100 patients with stable angina or NSTEMI (non-ST-segment elevation myocardial infarction) demonstrated that vFFR correlated well with FFR (r=0.89, p<0.001). The feasibility and diagnostic performance of vFFR in the catheterization laboratory remains unknown. In 231 vessels and 199 patients vFFR were evaluated by an expert (>100 cases performed) and a non-expert (<20 cases performed). Analyses showed operator variability of vFFR with results influenced by operator experience of vFFR processing. These results highlight the importance of training and quality assurance to ensure reliable, repeatable vFFR results [60].

Fractional flow reserve derived from intracoronary imaging

Intracoronary imaging techniques such as optical coherence tomography (OCT) and intravascular ultrasound (IVUS) provide high resolution and accurate 3D vessel reconstructions. Intracoronary imaging derived approaches are less developed than ICA-derived approaches, partly due to the lower availability and experience with intravascular imaging. Existing studies presented promising diagnostic accuracy estimates approaching 90% with FFR as reference standard ( Table 1) [61, 62, 63, 64, 65, 66]. Compared with ICA-based approaches, physiology derived from intracoronary imaging appears to be more reproducible which is most likely explained by the higher spatial resolution securing a more accurate geometrical reconstruction less influenced by manual adjustments.

Fractional flow reserve derived from OCT

CFD-based methods are available for derivation of FFR from OCT imaging. Ha et al. reconstructed 3D vessel geometry with no side branches from OCT and applied steady-state CFD for FFR computation (FFROCT) [61] ( Table 1). Generic patient-averaged flow velocity was applied at the inlet boundary of the reconstructed vessel, and averaged measured blood pressure was applied at the outlet boundary. Steady-state CFD analysis was used subsequently to solve the fluid dynamics equations governing blood flow. The average total analysis time for FFROCT was less than 10 minutes. A retrospective single-center study investigating 92 patients with intermediate stenosis in LAD demonstrated that the diagnostic accuracy, sensitivity and specificity of FFROCT was 88%, 68.7% and 95.6%, respectively ( Table 1). However, the discriminative power (AUC=0.93) was not superior to minimal lumen area (AUC=0.93). Repeatability analysis showed intra- and inter-observer variability of 0.01±0.05 and 0.01±0.01, respectively. The diagnostic performance of FFROCT in non-LAD lesions has not been validated. Furthermore, the generic one-size-fits-all flow model may not be applicable to all patient subsets. In addition, side branches are neglected in FFROCT model to simplify the CFD solution. The latter can result in underestimation of FFR [67].

Lee et al. introduced an approach that computes FFR based on OCT-derived morphology coupled with a lumped model of the coronary microcirculation (OCT-FFR) [62] ( Table 1). Hyperaemic coronary blood flow is estimated from measured stroke volume, heart rate and assumed microvascular resistance reduction during hyperaemia. The flow distribution to each major vessel is estimated based on the total length of main and sub-vessels. The downstream microcirculation is represented by a simplified model and is coupled to the reconstructed vessel of interest [68]. The average computational time (not including manual adjustment) for OCT-FFR was approximately 29 minutes. In a small population (17 vessels from 13 patients), OCT-FFR showed good correlation and agreement with FFR, and demonstrated a diagnostic agreement of 94% with FFR [62] ( Table 1). Side branches are neglected with OCT-FFR.

Seike et al. uses basic fluid dynamic principles to derive FFR based on OCT vessel morphology and patient-specific stenosis flow reserve (SFR) within ten minutes [69] ( Table 1). Pressure drop over the stenosis is calculated from its quadratic relation with hyperaemic flow, which causes both viscous friction and exit separation as blood flows through the stenotic segment. By estimating the hyperaemic flow velocity from patient-specific SFR, pressure drop over the stenosis is obtained. The validation study showed that FFR computed by this method correlated better with pressure wire-based FFR (r=0.89) than percent diameter stenosis measured by QCA (r=-0.65) and minimal lumen area (r=0.68) measured by OCT. The diagnostic performance of this method in a clinical setting has not been validated.

Yu et al. recently extended the QFR computational algorithm to derive FFR from OCT images and developed a fast method to compute OCT-based FFR (OFR) [63] ( Figure 1 & Table 1). The lumen contours are automatically delineated and stacked in 3D for each OCT image pullback. Subsequently, the cut-planes of the side branches ostia are reconstructed to quantify the areas of the side branches. Based on bifurcation fractal laws and the step-down phenomenon at the coronary bifurcations, the reference lumen as if there was no stenosis is reconstructed. Finally, the size of the reference lumen is multiplied with a fixed flow velocity of 0.35 m/s to derive the hypothetic volumetric flow rate and used as the boundary conditions for the QFR algorithm to compute FFR at every position along the reconstructed vessel. Specifically, for interrogated vessels with stents, stent struts are automatically detected and combined with the lumen geometry to compute OFR pullback along the stented segment [70]. Stent apposition and stent expansion can be quantified based on the 3D reconstruction of the stent struts [71].

In a retrospective international multicentre study investigating 125 vessels from 118 patients the diagnostic accuracy, sensitivity and specificity of OFR were 90%, 87% and 92% with FFR as reference [63] ( Table 1). The discriminative power of OFR was significantly higher than minimal lumen area measured by OCT (AUC = 93 versus 0.80, p = 0.002). OFR was reproducible with inter- and intra-observer variability of 0.00 ± 0.03 and 0.00 ± 0.02. The average total analysis time for OFR was 55 ± 23 seconds illustrating the potential for real-time physiological evaluation in the catheterization laboratory. A prospective multicentre study enrolling 60 patients with 76 vessels demonstrated high in-procedure feasibility (98.7%) and reproducibility (ICCa = 0.97 for intra- and 0.95 for inter-observer reproducibility) of OFR in a real-world series [65] ( Table 1). Different OCT pullback speeds had negligible influence on OFR computation, provided the image pullbacks were acquired on the same coronary segment. In a retrospective study comparing OFR and QFR in de novo lesions and in-stent restenosis from 181 patients with 212 vessels, OFR showed significant better correlation with FFR than QFR (r = 0.87 versus 0.77, p<0.001) and superior diagnostic performance (AUC = 0.97 versus 0.94, P = 0.017) [64]( Table 1). The diagnostic accuracy of OFR remained statistically equivalent by presence of implanted stents or previous myocardial infarction. In 103 coronary arteries undergoing successful PCI with a stent, OFR immediately after PCI also showed high correlation with FFR in both the entire vessel (r = 0.84, p<0.001) and the stented segment (r = 0.69, p<0.001) [72]. A recent post-hoc analysis of combined datasets from the DOCTORS study and OxOPT-PCI study showed good diagnostic concordance (84%) between post-PCI OFR and post-PCI FFR [70]. Stent underexpansion was significantly correlated with in-stent pressure drop derived from OFR pullback [70].

Fractional flow reserve derived from IVUS

The feasibility of computational FFR derived from IVUS imaging was first reported by Carrizo et al [73]. Seike et al. extended their OCT-based FFR method to IVUS imaging (IVUS-FFR) [74]. All the assumptions and fluid dynamic equations are unaltered, except that IVUS-derived morphological parameters are used to estimate pressure loss caused by viscous friction. In a validation study, IVUS-FFR correlated better with invasive wire-based FFR than minimal lumen area measured by IVUS (r = 0.78 versus r = 0.43, p = 0.002) [74].

Bezerra et al. developed a CFD-based solution based on IVUS-derived vessel reconstructions [75]. Segmented lumen contours of the main vessel are positioned along the transducer trajectory reconstructed from two orthogonal angiographic images. Baseline coronary blood flow is derived from cardiac output, heart rate, age, and weight using a literature-validated formula. The inlet hyperaemic flow of the interrogated artery is estimated based on scaling laws and a predictable reduction of microvascular resistance during hyperaemia. Resistance downstream of the reconstructed vessel is estimated from scaling laws relating flow to tissue volume and is applied at the outlet boundary. A steady-state CFD solution is applied for calculation of FFR. The average computational time for IVUSFR was approximately 72 minutes. A small prospective study demonstrated diagnostic accuracy, sensitivity and specificity of 91%, 89% and 92% with FFR as reference ( Table 1). However, the study was limited by its small sample (33 vessels of 24 patients) and relatively low disease prevalence (27% with FFR below 0.80). The long analysis time for IVUSFR limits its in-procedure applicability in the catheterization laboratory.

Ultrasonic flow ratio (UFR) is a recently developed novel method for computation of FFR from IVUS images, based on the previous validated QFR computational algorithm [66] ( Figure 1 & Table 1). The computation of UFR is similar to that of OFR, consisting of vessel geometry reconstruction and pressure drop calculation. Both the lumen contours and the external elastic lamina (EEL) were automatically delineated using a deep learning-based algorithm and reconstructed in 3D from IVUS pullback images. Subsequently, the healthy reference lumen size was calculated considering the step-down phenomenon at the coronary bifurcation, and combined with the pre-specified boundary conditions for the QFR algorithm to compute FFR along the reconstructed vessel.

In a retrospective single-center study investigating 97 vessels from 94 patients with 167 paired comparisons (79 before and 88 after PCI) between UFR and FFR, the diagnostic accuracy, sensitivity, and specificity of OFR was 92%, 91%, and 96%, respectively [66] ( Table 1). UFR showed significantly higher discriminative ability than IVUS-derived MLA (AUC = 0.97 versus 0.89, p<0.001). The repeatability of UFR was high with inter- and intra-observer variability of 0.01 ± 0.03 and 0.00 ± 0.03, respectively. Median total analysis time for UFR was 102 (IQR: 87 to 122) seconds, indicating a promising tool for real-time physiological evaluation in the catheterization laboratory. The in-procedure feasibility and diagnostic performance of UFR remain to be assessed.

FOCUS BOX 1 - The concept of FFR computation from coronary imaging
  • Reconstruction of geometrical data based on non-invasive or invasive coronary imaging
  • Defining hemodynamic boundary conditions
  • Computation of pressure data using computational fluid dynamics or fluid dynamics equations
  • Naturely build-in co-registration between anatomy and physiology
  • Possibility to associate plaque characteristics with functional evaluation
FOCUS BOX 2 - Quality image acquisition to obtain optimal FFR computation
  • Nitrates administration prior to image acquisition is mandatory
  • No other hyperemic agent is needed
  • FFR computation derived from coronary computed tomography angiography acquisition should include the entire heart, be performed according to a protocol to lower the heart rate without motion or misalignment
  • FFR computation derived from invasive coronary angiogram require at least two different angiographic views without overlap, foreshortening, panning and with long, brisk contrast injections
  • FFR computation derived from intracoronary imaging should cover all the lesions in the interrogated vessels

Application of non-invasive physiology in clinical practice

Improved gate-keeper function to the catheterization laboratory

The role of CCTA and CCTA + FFRCT as gatekeeper is revisited following the ISCHEMIA trial that showed no benefit of an early invasive strategy in patients with CAD (>50% DS in major epicardial artery on CCTA and moderate-severe ischemia on non-invasive stress testing [76]. It is appealing to increase the use of CCTA in the light of ISCHEMIA because CCTA is excellent to rule-out CAD and high-risk anatomy (e.g. Left main disease) that were not eligible for enrolment in ISCHEMIA. Furthermore, functional testing (e.g. FFRCT) on top of CCTA may be beneficial to identify lesion-specific ischemia in patients not controlled by medical therapy alone and in patients with a high-risk profile (e.g. low left ventricular ejection fraction, severe symptoms and severe abnormal imaging) [77]. The growing body of evidence clearly suggests that FFR derived from coronary CTA has the potential to become a game-changer as gate-keeper to the catheterization laboratory. The introduction of FFRCT has improved the diagnostic accuracy of CCTA. CCTA in combination with FFRCT can prevent unnecessary invasive procedures and revascularization in patients with a low risk profile. Several centers have implemented CCTA with FFRCT analysis as part of their daily clinical workflow [78, 79]. According to the NICE guidelines, FFRCT should be considered as an option for patient with stable chest pain as it leads to substantial cost reduction by avoiding invasive investigations and unnecessary treatment. Clinicians should however acknowledge the importance of appropriate patient selection as current validation studies on FFR CT mainly included patients already scheduled to undergo ICA [18, 19, 20]. The clinical value and economic efficiency of applying FFRCT in patients with a low pre-test risk of coronary artery disease is thus unclear. Further, despite that the average turnover time for receiving the clinical results has improved substantially, the work-stream is still cumbersome as the process requires upload of images to a remote core laboratory for analysis. The same concept of optimized referral applies to diagnostic catheterization laboratories not able to or not allowed to perform FFR [80].

There are several additional limitations that should be acknowledged. First, artefacts related to CCTA quality including calcifications and cardiac motion affect the feasibility. Limitations related to low temporal/spatial resolution and image noise may give false-positive results. Thick CT slice thickness and high heart rate are predictors of insufficient quality for FFRCT analysis [81]. In the DeFACTO and NXT studies, 11% and 13% of cases were rejected because of low image quality. Few studies account for the lower feasibility rate of FFRCT in head-to-head comparisons with other non-invasive modalities such as perfusion imaging (e.g. with intention-to-diagnose analysis). Secondly, in a study comparing FFRCT to invasive FFR for staged evaluation of non-culprit lesions in STEMI patients showed lower diagnostic accuracy of FFRCT, probably due to impaired myocardial mass hindering accurate estimation of coronary blood flow [82]. Therefore, careful interpretation should be made in patients with prior myocardial infarction or suspected impaired left ventricular myocardial mass. Additionally, the use of FFRCT is not validated in vessels previously revascularized by PCI or coronary artery bypass graft. Finally, despite that the average bias of FFRCT may seem acceptable, the measurement uncertainty exceeds 10% (standard deviation to the FFR-FFRCT difference >0.10) [83, 84].

Revascularization decision making

Once the patient enters the catheter laboratory, appropriate identification of lesions requiring revascularization is of foremost importance. The described advances in technologies have led to hypotheses that a physiological roadmap may be helpful in order to guide and optimize subsequent revascularization. Given by the broad 95 % accuracy limits of FFRCT, further physiological assessment may be necessary even if a CCTA derived physiological roadmap is available [83]. In patients with three-vessel disease, physiology-guided PCI using wire-based approaches improved the clinical outcome while treating fewer lesions when compared to conventional PCI [85]. Calculation of functional SYNTAX score (FSS) provides a better discriminative ability to predict adverse events compared to the anatomical SYNTAX score [86]. CCTA with FFR CT and ICA with QFR showed good FSS agreement compared with FSS calculated with FFR [42, 87]. In heart teams, treatment planning based on CCTA combined with FFR CT agrees with treatment planning based on ICA in roundly 80% of the cases [88].

Besides the functional information, ICA-derived applications further provide detailed anatomical information applicable for stent sizing, derivation of optimal viewing angles and co-registration to intracoronary imaging [89, 90, 91, 92, 93]. In addition, optimal fluoroscopic viewing angles can accurately be derived from coronary CTA in order to optimize the invasive visualization of coronary ostia and bifurcation lesions which may be of particular interest for planning of the invasive procedure [94].

PCI optimization

Despite angiographically successful PCI, recent registries demonstrated high incidence of suboptimal (FFR ≤0.90) or frankly impaired (FFR ≤0.80) coronary physiology [95, 96]. Low FFR value is associated with worse outcome [15, 97, 98, 99]. Suboptimal physiology immediately after PCI can be caused by multiple factors, including stent underexpansion, stent malapposition, tissue protrusion and edge dissection, and untreated disease outside the stent [15]. Imaging-derived physiology could constitute a role in post-PCI physiological lesion assessment because of the good pre-stenting agreement with FFR. As pure anatomical index, IVUS-guided PCI improves clinical outcome and the same could be expected for OCT [100]. Given the higher spatial resolution when compared to angiography, the combination of intracoronary imaging derived physiology and anatomy could further improve clinical outcome and will be subject to future studies. The use of OFR and UFR that includes the actual stent apposition and expansion information in the computational FFR might be suitable for optimization of PCI during the procedure ( Figure 2)[63, 66].

Virtual stenting

Detailed anatomical information derived from CCTA with physiological information derived from FFRCT provide a treatment-planner tool that is currently being tested in a prospective international study for agreement with post-PCI FFR [101, 102, 103]. Likewise, Gosling et al. introduced a method to predict post-PCI FFR before stenting with computational vFFR derived from ICA [104]. Recently, post-hoc analysis of DOCTORS study validated simulated residual QFR, which is calculated from pre-PCI QFR with virtual stenting, in predicting post-PCI QFR and FFR, indicating a promising tool in predicting the stenting result prior to proceeding with stent implantation [105]. Furthermore, case series performed in two patients with sequential stenoses measured with FFR and OCT-derived FFR showed the feasibility of OCT-derived FFR to predict the FFR value post stenting of one stenosis [106]. The recently proposed simulated residual OFR, a method for calculation of post-PCI OFR from pre-PCI OCT run after elimination of the stenotic segment by virtual stenting, was shown significantly correlated with post-PCI FFR [70]. The non-invasive virtual stenting tools are of particular interest in sequential stenosis where wire-based physiology results are challenging to interpret because of inter-stenosis cross talk [107].

STEMI patients with multivessel disease

The use of non-invasive physiology may be beneficial in STEMI patients with multivessel disease by identifying non-culprit lesions that benefit from revascularization as it may reduce the procedure time and potentially spare unnecessary down-stream procedures [43, 51, 82]. If aiming for complete revascularization during the index procedure, CCTA with FFRCT does currently not appear feasible given the long turn-over time for completed analysis. However, it could be used for planning of a staged revascularization procedure although initial data showed that the diagnostic performance of FFRCT is solely moderate for staged detection of ischemia in STEMI patients with multivessel disease [82]. It should however be acknowledged that the validity of FFR measurements in NCL is questionable as multiple reports document altered microvascular conditions even in a staged setting [108]. For ICA derived computational approaches, a proof of concept study showed that QFR-examined incomplete functional revascularization was associated to a higher incidence of a patient orientated composite endpoint than complete functional revascularization (Hazard ratio 2.3 (95% CI:1.2-4.5), p=0.01) [51]. An advantage of ICA and intracoronary imaging derived modalities over CCTA-based approaches is the in-procedure feasibility allowing for complete revascularization during the index procedure. The optimal revascularization strategy for STEMI patients with multivessel disease is a subject of current debate but regardless of the outcome computational physiology can potentially be valuable in this setting.

FOCUS BOX 3 - Clinical applications of non-invasive physiological assessment
  • Improved referral to ICA of patients with suspected CAD
  • Fast and safe access to a physiological-guided PCI in patients with intermediate stenosis
  • Virtual PCI-planning in patients with multivessel disease and sequential coronary artery stenosis
  • Post-PCI optimization using combined physiology and high resolution intracoronary imaging
  • Evaluation of non-culprit lesions in STEMI patients with multivessel disease

Future perspectives

Reference standard for assessing the ischemic potential of epicardial disease

FFR became the reference standard to assess the functional impact of coronary stenosis and to predict the clinical results achieved with revascularization. However, in several situations FFR showed some limitations such as for evaluation of sequential stenosis [109], tortuous vessels generating accordion phenomenon by the pressure wire [110] or stenosis evaluation of a vessel supplying a chronic total occlusion area [111]. Computational FFR measurements could theoretically solve these situations ( Figure 3). In diffuse and serial lesions where FFR evaluation remains challenging even with cautious pullback under intravenous infusion of adenosine, computational FFR seems to show an interest. This aspect has been studied with a non-invasive FFRCT-based PCI planner tool to predict the FFR contribution of each stenosis in serial lesions [103] and with OCT-derived FFR to predict the FFR value post stenting of one stenosis [106] in few patients. Thus, computational FFR has the potential to be the reference standard in these situations, since it is safer, cheaper and most of the techniques derived from ICA are faster than FFR with high diagnostic accuracy.

Increasing the adoption of physiology-guided PCI

Despite the increasing accumulation of clinical data and guideline recommendation supporting the use of invasive coronary physiology, the global FFR adoption map in 2016 illustrates a heterogeneous adoption with numerous areas using FFR in <6% of ICA’s [9]. Numerous arguments could explain the low overall adoption including wire-cost, discomfort and reimbursement systems not favouring FFR. However, recent data suggests operator’s confidence in eyeballing remains the most important struggle to overcome [112].

Little doubt exists that computational FFR bears the potential to increase the use of physiology-guided PCI in a broader population worldwide. We list the pros and cons of each technique to calculate FFR in a comparative table ( Table 2). Currently, pending clinical outcome studies, computed physiology can be used in a hybrid approach where wire-based FFR is used in small range of values around the diagnostic cut-off value. The latter process results in wire- and adenosine free procedures in 2/3 of all procedures while ensuring a high classification agreement with FFR [45]. We reported in the Figure 4 main studies comparing non-invasive stress test or imaging to FFR as gold standard. We can appreciate in this figure how computational FFR techniques have close diagnostic accuracies with respect to FFR and most sophisticated non-invasive imaging modalities such as cardiac magnetic resonance imaging or cardiac positron-emission tomography.

Combination of plaque characteristics and physiology derived from coronary imaging

CCTA derived plaque characteristics (e.g. non-calcified plaque volume, low attenuation plaques) are inversely related to FFR [113, 114]. Fluid structure interaction methods simulating the blood flow in ideal stenotic coronary arteries with different plaque components found that FFR values tend to be higher in deformable lipid-rich plaques than in rigid fibrous and calcified plaques. Nevertheless, plaque composition was not associated with FFR after adjusting for other parameters including plaque burden and lesion length [115]. This finding justifies the current FFR computation principles that do not take into account the plaque characteristics. However, there is unquestionably room to refine the diagnostic strategy for patient with stable angina and coronary artery disease. The recently published 5-years results of FAME II demonstrate that FFR continuous to be superior to angiography driven by a lower rate of urgent revascularization [5]. Albeit, functional assessment of coronary stenosis to predict outcome could be affine with plaque burden and composition from imaging (CCTA or intravascular imaging) because cardiovascular event do also occur in patients without apparent obstructive CAD on CCTA [116]. Thus, it may be favourable to incorporate other prognostic factors contributing the nature of CAD. Lee et al. showed that high risk plaque characteristics have a significant association with the risk of clinical event even with an FFRCT value>0.80 [117]. A recently developed automatic framework for plaque characterization from OCT images using artificial intelligence showed excellent agreement (r2 = 0.98) and good diagnostic accuracy (98%, 91%, and 89% for fibrous, lipid, and calcium plaques) with standard manual measurements [118]. Automatic detection of high-risk plaque features combined with computed physiology could refine the diagnostic strategy and will be subject to future studies.

Furthermore, the potential to derive physiology from intravascular imaging is to obtain the functional impact of a lesion of interest on top of the morphological information. Intracoronary imaging approaches can accurately identify high-risk plaques (e.g. thin-cap fibroartheromas) associated to MACE [119]. It will allow to reduce the cost of the procedure and combine detailed plaque characteristics, nature of the lesion, differentiate plaque rupture or erosion, dissection or thrombus with functional impact without the additional cost of a pressure wire.

Outcome-trials

The presented validation data justifies the rationale for conducting randomized trials that will inform on the applicability of computational physiology to guide clinical decision-making.

The randomized FORECAST trial (NCT03187639) is assessing whether FFRCT is superior to routine clinical pathways for patients with stable chest pain as recommended by the NICE guidelines. The primary endpoint is resource utilisation. The randomized FAVOR III China trial (NCT03656848) will address whether a QFR-guided revascularization strategy yields superior clinical outcome and cost-effectiveness compared to angiography-guided revascularization strategy[120]. The primary endpoint is MACE consisting of all-cause mortality, any myocardial infarction and any ischemia-driven revascularization. The randomized FAVOR III Europe-Japan trial (NCT03729739) assesses whether a QFR-guided revascularization strategy is non-inferior to an FFR-guided revascularization strategy on a patient oriented composite endpoint consisting of all-cause mortality, any myocardial infarction and any unplanned revascularization.

FOCUS BOX 4 - Take-home messages
  • Pure anatomical assessment of epicardial coronary artery disease is insufficient to guide coronary revascularization decision making
  • Advanced fluid modeling can be applied to three dimensional geometrical reconstructions to identify flow limiting epicardial coronary artery disease
  • Computational physiology derived from coronary computed tomography, invasive coronary angiography or intracoronary imaging has a high diagnostic performance with fractional flow reserve as reference standard
  • Randomized clinical outcome data is awaited to document the clinical applicability of computational physiology to guide PCI

Conclusion

Combined imaging and fluid dynamics allow for computation of non-invasive FFR from coronary imaging such as CCTA, ICA and intravascular imaging. Due to minimal requirements for data acquisition, no use of drugs and pressure devices, computational non-invasive FFR derived from ICA can be straightforwardly incorporated into routine clinical practice with the potential to improve the adoption of physiological guidance of coronary revascularisation.

Personal perspective – Shengxian Tu

Fractional flow reserve (FFR) is the present standard method for functional assessment of epicardial coronary stenosis during invasive coronary angiography. The combined morphological and functional evaluation of coronary stenosis emerges as the standard of care in the workup of patients with known or suspected coronary artery disease (CAD). This “co-registration” between anatomy and physiology has the potential to become the cornerstone of clinical decision making for appropriate selection or medical therapy, PCI, bypass surgery or hybrid approach. In the past years, the use of pressure wire-based evaluation has increased but remains overall heterogeneous and very low in most healthcare systems. To improve the clinical adoption of physiological guidance of coronary revascularization, several computational approaches were developed to derive FFR from imaging data without the need to use pressure wire or induce hyperemia. CT-based FFR opens the opportunity for low to intermediate risk patients with suspected CAD to undergo “one-stop shop” evaluation of coronary anatomy and physiology prior to referral for invasive coronary angiography. Angiography-based FFR enables fast computation of FFR during coronary angiography with simultaneous generation of accurate 3D quantitative data for optimization of stent sizing and positioning. Finally, intracoronary imaging-based FFR has great advantage in computational efficiency and reproducibility while allowing for detailed assessment of plaque morphology and interventional devices. Awaiting outcome data from several on-going randomized controlled clinical trials, computational physiology might be widely incorporated into routine clinical practice for improved identification of flow-limiting coronary stenosis, optimization of revascularization strategies, and evaluation of the interventional devices.

SHARE YOUR COMMENT