Noninvasive CT-based hemodynamic assessment of coronary lesions derived from fast computational analysis: a comparison against fractional flow reserve

Application of computational fluid dynamics (CFD) to three-dimensional CTCA datasets has been shown to provide accurate assessment of the hemodynamic significance of a coronary lesion. We aim to test the feasibility of calculating a novel CTCA-based virtual functional assessment index (vFAI) of coronary stenoses > 30% and ≤ 90% by using an automated in-house-developed software and to evaluate its efficacy as compared to the invasively measured fractional flow reserve (FFR). In 63 patients with chest pain symptoms and intermediate (20–90%) pre-test likelihood of coronary artery disease undergoing CTCA and invasive coronary angiography with FFR measurement, vFAI calculations were performed after 3D reconstruction of the coronary vessels and flow simulations using the finite element method. A total of 74 vessels were analyzed. Mean CTCA processing time was 25(± 10) min. There was a strong correlation between vFAI and FFR, (R = 0.93, p < 0.001) and a very good agreement between the two parameters by the Bland–Altman method of analysis. The mean difference of measurements from the two methods was 0.03 (SD = 0.033), indicating a small systematic overestimation of the FFR by vFAI. Using a receiver-operating characteristic curve analysis, the optimal vFAI cutoff value for identifying an FFR threshold of ≤ 0.8 was ≤ 0.82 (95% CI 0.81 to 0.88). vFAI can be effectively derived from the application of computational fluid dynamics to three-dimensional CTCA datasets. In patients with coronary stenosis severity > 30% and ≤ 90%, vFAI performs well against FFR and may efficiently distinguish between hemodynamically significant from non-significant lesions. Virtual functional assessment index (vFAI) can be effectively derived from 3D CTCA datasets. In patients with coronary stenoses severity > 30% and ≤ 90%, vFAI performs well against FFR. vFAI may efficiently distinguish between functionally significant from non-significant lesions. Virtual functional assessment index (vFAI) can be effectively derived from 3D CTCA datasets. In patients with coronary stenoses severity > 30% and ≤ 90%, vFAI performs well against FFR. vFAI may efficiently distinguish between functionally significant from non-significant lesions.


INTRODUCTION
Fractional flow reserve (FFR) has evolved as the invasive reference standard for functional assessment of coronary stenoses and guidelines recommend its routine application in patients undergoing ICA in the absence of previously non-invasively documented significant inducible ischemia [1]. In patients with stable CAD, FFR does not substitute the assessment of inducible myocardial ischemia, but it has proven to be superior to classical coronary anatomic parameters (i.e. stenosis severity) in guiding patients' management, reliably discriminating the coronary lesions which have hemodynamic significance and could be revascularized from those that can be safely managed conservatively, with improved patients' clinical outcome and overall resource utilization [1][2][3]. Nevertheless, FFR is not routinely performed in daily clinical practice, being used in less than 10% of the coronary diagnostic or interventional procedures [4], possibly due to added technical needs (i.e. need for a dedicated wire and administration of a coronary vasodilator) and higher cost (8).
In the last decade, computed tomography coronary angiography (CTCA) has become a valid diagnostic tool in different patient groups, allowing accurate non-invasive evaluation of CAD extent and severity [5,6]. Moreover, by application of computational fluid dynamics (CFD) to coronary three-dimensional anatomical models, it has been demonstrated that CTCA allows also non-invasive assessment of the hemodynamic relevance of a coronary lesion with good diagnostic accuracy compared to invasive FFR [7][8][9][10][11][12][13]. A number of recent studies have suggested that quantitation of CTCA-based FFR (FFRCT -HeartFlow, Inc., Redwood City, CA) might significantly increase the diagnostic accuracy of CTCA [14][15][16][17][18][19][20][21]. However, the current version of this technique requires a remote and lengthy core-laboratory analysis which could limit its wider clinical application. Kruk et al., proposed an alternative method using a software research prototype (cFFR version 1.4, Siemens AG Healthcare, Forchheim, Germany), which does not require a remote core-laboratory analysis [9,22,23], while Ko et al developed a method which still requires a core-laboratory analysis (Canon Toshiba Medical Systems Corp.), but is fast with a computational time around 30 minutes [13]. 4 The evaluation of virtual functional assessment index (vFAI) measured from invasive angiographic data [24] has been recently suggested as a valid alternative to FFR measurements in patients submitted to ICA, allowing to determine the hemodynamic relevance of a given coronary lesion with a few minutes long computation time. The algorithm uses three-dimensional (3D) coronary anatomical data and steady-flow CFD analysis to compute the ratio of distal to proximal pressure over the lesion for flows in the range of 0 to 4 ml/s, normalized by the ratio over this range for a normal artery, offering a measure of CAD hemodynamic significance that is numerically equal to the invasively measured FFR.
However, while vFAI measures have been successfully derived from 3D-ICA datasets, positively mimicking invasive FFR results, the possibility to compute this coronary functional parameter from CTCA-based coronary anatomical models has not yet been investigated. The first aim of the current study is to test the feasibility of assessing vFAI by using our automated in-house developed CTCA 3D reconstruction software for generating 3D arterial models, and performing the required blood flow simulations on the aforementioned models [25]. The second aim is to evaluate the efficacy of our method by comparing the derived vFAI values to the invasively measured FFR in coronary stenoses ranging between 30%-90%.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 Patients with previous acute coronary syndrome, known CAD, left ventricular ejection fraction <35%, more than moderate valve disease, and cardiomyopathy were excluded. From the enrolled patients, those with From this group, patients with satisfactory-to-excellent image quality of CTCA acquisition as defined by the EVINCI CTCA core-lab, based on absence of motion artifact, image noise or extensive coronary calcifications, were selected for evaluating the accuracy of vFAI compared to invasive FFR to predict the presence of hemodynamically significant CAD. Accordingly, from the 69 patients, 6 (8.7%) were excluded from the study due to poor image quality.

Invasive coronary angiography and FFR measurements
ICA was performed using standard techniques and multiple projections. Quantitative analysis by ICA core-lab was available in all the selected patients. The FFR was invasively measured using a Combo-Wire (Volcano Therapeutics, Rancho Cordova, California), under the administration of 140 μg/kg/min of intravenous adenosine. Segments with FFR values ≤0.8 were considered to indicate significant stenoses.

CTCA acquisition protocol
Coronary CTCA was performed using ≥64-slice CT scanners Coronary CT angiography was performed in all of the enrolled patients, as mandated by the original protocol of the EVINCI study [26], using ≥64-slice CT scanners.. The arterial segments of interest were reconstructed in mid to end diastole (70%-80% of the R-R interval) with an average slice thickness of 0.6 mm and an increment of 0.6 mm. To optimize image acquisition and final CTCA quality, beta-blockers and sub-lingual nitrates were used as per study protocol. In order to improve the interpretability of CT images and maximize the use of dose-sparing prospective ECG-gating, a regular heart rate of <65 bpm was adopted by the study protocol. Accordingly, the use of beta-blockers was encouraged in case of higher heart rates, by means of up to 25 mg of intravenous metoprolol.  The presence of coronary calcifications was evaluated in each patient by computation of calcium score (CACS), which was considered extensive if it was higher than 400 [26]. Scan quality of the CTCA was categorized by the independent EVINCI Core-Lab into 4 categories (i.e. excellent, good, satisfactory and poor). Scans of poor quality were excluded from the present analysis.

CTCA 3D Reconstruction
The 3D reconstruction was performed using our in-house developed software and the reconstruction process is carried out in 7 steps, as it is described elsewhere [25].
Coronary arteries were reconstructed using the same landmarks for each patient based on the SYNTAX SCORE chart. Specifically, for RCA, the reconstructed models included segments 1-3, for the LAD artery, segments 6-8, whereas for the LCx artery, segments 11-13. The reconstruction and the vFAI calculation process are depicted in Fig. 1

Virtual Functional Assessment Index (vFAI) calculation
Blood flow simulations were performed on the 3D models of the arteries by generating a mesh of tetrahedral finite elements which then allows the solution of the appropriate Navier-Stokes and continuity equations using ANSYS® CFX 15 (Canonsburg, USA).
For the vFAI calculation, following the method of Papafaklis et al. [24], we performed two separate simulations for each case, applying flow rates of 1 and 3 ml/s, in order to create the case-specific pressure gradient (ΔP)-flow relationship [27] for each case (Eq. 1): where ΔP is the pressure gradient, Q is the flow rate, fν is the coefficient of pressure loss due to viscous friction and fs is the coefficient of pressure loss due to flow separation [28][29][30]. By solving the fully determined system of the aforementioned equation, we calculated Pd/Pa from (Eq. 2): Pa was set at 100 mmHg (mean aortic pressure) , hence the area under the Pd/Pa vs. flow curve is 1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 then calculated for a flow range between 0 and 4 ml/s, which corresponds to the mean+2SD increase of the hyperaemic flow rate in a normal human coronary artery, initiating from an average flow rate value of 1 ml/s during rest [31]. Finally, we calculated the vFAI for each case as the ratio of the area under the artery-specific Pd/Pa vs. flow curve to the reference area, a value that has been shown to correlate well to the invasively measured FFR value [24]. Although vFAI was tested and validated on ICA derived 3D arterial models, the fluid dynamics background of the aforementioned method may also be applicable to CTCA derived models, since it takes into account only the

Statistical analysis
The relationship between FFR and vFAI was quantified by calculating the Pearson's correlation coefficient. Bland-Altman plots and the corresponding 95% limits of agreement were used to assess the agreement between the two methods. ROC analysis was performed for identifying the cut-off values of the examined variables. Categorization of FFR and vFAI values was made using the cutoff of 0.8 and the calculated cut-off from ROC curve for the FFR and vFAI, respectively. ANOVA was used for multiple group comparisons followed by Bonferroni correction where applicable.  1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 and diagnostic accuracy (the percentage of patients correctly diagnosed by vFAI) were used to assess the performance of vFAI. P values <0.05 were considered statistically significant.
Comparison ROC curves was based on the DeLong method (MedCalc software). The McNemar test was used to compare the accuracy of vFAI and CCTA to predict ischemia causing stenosis. P values <0.05 were considered statistically significant.

Baseline characteristics of the study population
The population of the present study included 63 patients (mean age 62.9 ± 7 years, 61 % males) with stable symptoms. Hypercholesterolemia, hypertension and CAC score > 100 were present in more than 60% of patients. The mean heart rate during the CTCA scans was 60 ± 7 beats/min.
Patients' characteristics are summarized in Table 1

CTCA derived vFAI measurements and FFR
CTCA scan quality was defined as good-excellent for 65 (87.8%) coronary arteries and satisfactory for 9 (12.2%) vessels. The interobserver agreement for vFAI measurements was tested in 11 randomly selected coronary vessels (4 RCA, 7 LAD) (8 of good-excellent and 3 of satisfactory quality There was a strong correlation between vFAI and FFR (R=0.93, p<0.001) and a very good agreement between the two parameters by the Bland-Altman method of analysis (Fig. 2). The mean difference of measurements was 0.03 (SD=0.033, p<0.0001), indicating a small systematic overestimation of the FFR by vFAI. The corresponding limits of agreement were from -0.03626 to 0.09599, with 95% confidence intervals of -0.04968 to -0.02284 for the lower limit and 0.08257 to 0.1094 for the upper limit. Agreement was unaffected by calcification level. For calcified vessels with Agatston Score>400, (n=23) the mean difference was 0.036±0.065, whereas for milder calcifications (Agatston Score<400, n=51), the mean difference was 0.036±0.047.
A number of studies have already reported the ability of CTCA derived functional indexes to mimic invasive FFR measurements for the detection of hemodynamically significant CAD [7,17,33]. In particular, while initial reports had suggested the existence of only a moderate agreement between FFRCT and invasive FFR [12], recent refinements of computational techniques and CFD modelization have improved the accuracy of FFRCT in unmasking the presence of hemodynamically significant coronary lesions and quantifying their severity [33]. Our results show that vFAI slightly overestimates FFR but can distinguish functionally significant from nonsignificant lesions with high specificity and NPV. In our study population, the cut off vFAI value of 0.8, which has been used in prior studies, would have resulted in a low sensitivity (66.7 %). Due to the slight overestimation of FFR by vFAI demonstrated in the present study, the best cut off value for vFAI was ≤0.82, which increased sensitivity to levels comparable to those of similar studies without affecting significantly the specificity. Our results also demonstrate that for studies of adequate quality, coronary calcification did not affect significantly the agreement between vFAI and FFR. Intense calcification affects image quality by creating the so-called "blooming effect" leading to an overestimation of the arterial lumen and thus restricting data analysis only to studies without obvious artefactual effects. We are currently working to resolve these issues by developing a dedicated "blooming removal" algorithm and further denoising filters .   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  +0.03) [7,9,17,33]. These studies use a different approach regarding the calculation of FFRCT, where the entire coronary vasculature is reconstructed in 3D and the lumped parameter models of the microcirculation to the outflow boundaries are coupled but at the expense of increased computational time and need of strong computational power. Using another approach, Taylor

Clinical implications
A preemptive non-invasive evaluation of the presence of hemodynamically significant CAD is favored by current clinical guidelines for avoiding unnecessary coronary interventions [35]. Whilst different cardiac imaging tests can be used to rule out the presence of functionally significant CAD, a significant proportion of patients currently submitted to ICA still show no hemodynamically significant coronary lesions [36]. Many of these patients have coronary atherosclerosis that can be depicted by CTCA. A combined assessment of coronary anatomy and function is therefore desirable, but it generally requires two separate diagnostic tests resulting in higher immediate costs. Accordingly, although the advantages of a comprehensive anatomo -1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 functional evaluation of patients with suspected ischemic heart disease have been confirmed [37,38], such an integrated assessment is rarely performed in daily routine.
While CTCA has long represented the reference standard for the non-invasive depiction of coronary anatomy, more recently it is also being used to obtain quantitative measures of CAD functional relevance through the modelization of coronary flow dynamics [7,39]. The calculation of FFRCT, the most validated of such functional parameters, typically requires the use of proprietary software with long computation times and a dedicated core-laboratory [17,33] [13]. Moreover, the proposed method requires minimal user interaction regarding the 3D reconstruction process of the desired arterial segment, resulting in a good interobserver as well as intraobserver agreement.

Study Limitations
We have performed a retrospective analysis in a rather modest sample size. have the ability to perform a full 3D reconstruction of the entire arterial tree and this will be tested in a larger prospective study aiming to include also individuals with clinical characteristics broader than those described in the current study. Finally, striving for an almost absolute agreement   1  2  3  4  5  6  7  8  9  10  11  12  13  14  15  16  17  18  19  20  21  22  23  24  25  26  27  28  29  30  31  32  33  34  35  36  37  38  39  40  41  42  43  44  45  46  47  48  49  50  51  52  53  54  55  56  57  58  59  60  61  62  63  64  65 between vFAI and FFR is probably an elusive task considering the intrinsic nature of vFAI. The latter mainly expresses the "potential" functional impact of anatomy, which differs fundamentally from the invasively calculated FFR that measures directly the effect of anatomy on real pathophysiologic conditions and thus, may be influenced by the vasodilating capability of the whole coronary system.