Non-invasive coronary physiology based on computational analysis of intracoronary transluminal attenuation gradient

Invasive procedure is a prerequisite for studying coronary physiology. We established the measurement of non-invasive physiological parameters including coronary blood flow (CBF), flow velocity, and microvascular resistance using coronary computed tomography angiography (CCTA). Vessel-specific CBF was derived from transluminal attenuation flow encoding (TAFE) and then tested using three separate datasets consisted of computational simulation, human perfusion CT, and human CCTA. TAFE-derived CBF correlated well with measured vessel-specific myocardial blood flow and CBF. TAFE-derived CBF per myocardial mass consistently decreased with the progressive severity of stenosis, and it was found to better to detect significant stenosis than transluminal attenuation gradient (TAG). With the addition of vessel anatomy, TAFE-derived CBF could calculate flow velocity and microvascular resistance. The results of non-invasively acquired parameters according to the severity of stenosis were similar to those obtained through invasive physiology studies. Our study demonstrated that non-invasive comprehensive coronary physiology parameters can be derived from CCTA without any pre-specified condition or performing complex heavy computational processes. Our findings are expected to expand the clinical coverage of CCTA and coronary physiology.

The key role of coronary artery is supplying sufficient blood flow which contains vital materials such as oxygen or glucose to match the needs of subtended myocardium. Therefore, quantification of vessel-specific coronary blood flow (CBF) has paramount importance in coronary physiology. However it is limited by requirement of invasive catheterization and use of dedicated intracoronary Doppler wire or infusion catheter 1,2 .
Coronary computed tomography angiography (CCTA) is a snapshot of dynamic intraluminal contrast transit. Therefore, intracoronary hemodynamics can be calculated by applying mass conservation and enhancement dynamics used in myocardial blood flow (MBF) quantitation 3 . Briefly, absolute CBF can be calculated from the time-dependent change of contrast density proximal to coronary artery as input function of contrast cohort, arterial volume to be filled by contrast cohort, and gradient of intraluminal contrast density that reflects flow velocity.
On the basis of this concept, Lardo et al. 4 have developed an elegant hemodynamics formula named transluminal attenuation flow encoding (TAFE) (Fig. 1). In this study, we investigated and validated TAFE using human subjects data. We validated TAFE using computational flow dynamics model and human perfusion CT. Further, we examined TAFE-derived CBF, flow velocity, and microvascular resistance in CCTA according to varying severity of coronary artery stenosis to determine the feasibility of non-invasive comprehensive coronary physiology studies.

Patients.
Between January 2011 and December 2014, we consecutively enrolled 30 perfusion CT cases without significant stenosis defined by diameter stenosis (DS) ≥ 50% and 100 CCTA cases with varying degrees of stenosis which were performed for elective evaluation of coronary artery disease. To minimize imaging bias related to temporal inhomogeneity, only CCTA acquired with single heart beat scan were enrolled. No patient was simultaneously enrolled in both studies. Patients with hypertrophic or secondary hypertrophy, prior myocardial infarction, total occlusion 5 , revascularization, heart failure, structural or congenital heart disease, prosthetic valves, or any clinical instability were not included. Institutional review board of Samsung Medical Center approved the study protocol dealing with anonymized image data; therefore, the requirement of informed consent was waived.

CT image acquisition. Second-generation dual-source CT scanner (SOMATOM Definition Flash; Siemens
Medical Solution) was used as previously described 6 . In perfusion CT, hyperemia was induced by continuous intravenous adenosine infusion of 140 μg/kg/min for 3 min. Intravenous contrast infusion included 70-80 ml of iomeprol (350 mg I/mL, Bracco) followed by 40 ml of saline at a rate of 4 to 5 ml/sec. Scan parameters were 280 msec gantry rotation, heart rate-dependent pitch 0.17-0.2, tube voltage 100 kV, and tube current 330 mA. Images were acquired for 30 sec during end inspiration. Dynamic datasets were acquired for every other R-R interval at two rapidly alternating table positions that provided 73 mm Z-axis coverage.
CCTA without hyperemic induction was performed with 2 × 128 × 0.6 mm sections and 2 × 64 × 0.6 mm detector collimation using Z-axis flying focal spot technique with prospective ECG-triggered high-pitch (pitch 3.2-3.4) helical mode. Contrast intravenous infusion consisted of 60 ml and then 40 ml of saline at a rate of 4 ml/ sec. Oral beta-blocker and nitroglycerin were administered and heart rate <60/min was confirmed.
Radiation-reduction technique (REDose4D, Siemens Healthcare) was applied as reasonably as possible. The mean effective radiation dose of perfusion CT and CCTA was 5.1 and 0.9 mSv, respectively.
CT Image reconstruction and measurement. A dedicated workstation (iNtuition, TeraRecon) was  Myocardial blood flow. In perfusion CT, LV-MBF was calculated using a dedicated volumetric perfusion analysis software (Leonardo, Siemens Medical) 6 . Then vessel-specific MBF was calculated by multiplying LV-MBF and the %fractional myocardial mass (%FMM), which is a vessel-specific myocardial mass on the basis of an allometric scaling between coronary artery length and myocardial mass 7,11 . In CCTA, resting vessel-specific MBF of normal coronary artery was assumed to be 0.90 ml/g/min 12 . Right coronary artery (RCA) supplies to both right ventricle (RV) and LV. RV was assumed to possess 36% of LV mass and the same MBF per myocardial mass (g) on the basis of the results of prior magnetic resonance imaging and pre-clinical studies that have demonstrated similar MBF per myocardium (g) in both ventricles 13,14 . Therefore, CBF of RCA proximal to the posterior descending artery bifurcation (PDA) and posterolateral artery (PL) was assumed to supply both RV mass and %FMM of LV mass subtended by PDA and PL.
Calibration and validation of TAFE formula. With the assumption that TAG is created by the advection of contrast dye into coronary artery from AIF, TAFE calculates vessel-specific CBF (ml/min) from average cross-sectional luminal area given as Â (cm 2 ), vessel length given as ŝ (cm), and time to peak enhancement given as T d (min) (Equation 1; Fig. 2A) 4 . Statistical analysis. Analyses were performed on a per-vessel basis unless specified. Categorical variables are presented as numbers and percents. Continuous variables were expressed as median with 1 st -3 rd quartiles in parentheses or as mean ± SD. TAFE coefficient k was determined by linear regression estimate between CBF and vessel-specific MBF. The relationship between CBF and MBF was tested by Pearson's correlation. Relationship among DS and physiologic parameters were interrogated with polynomial curve fitting using quadratic or cubic model, while the dose-response relation was evaluated by Jonchheere-Terpstra trend test. Two-tailed p-value < 0.05 was considered to be statistically significant. R version 3.4 (R foundation) was used. Patients. Both perfusion CT (n = 30) and CCTA (n = 99 after the exclusion of 1 case due to inadequate image quality) cases showed similar clinical characteristics (Table 1).
In summary, CBF per myocardium (g) decreased with progressive stenosis, although the flow velocity was maintained due to the concomitant reductions in luminal area, volume, and microvascular resistance (Fig. 4F). Interestingly, these curvilinear patterns of physiological parameters were similar to those of invasive physiology studies (Table 3 Non-invasively acquired parameters were additionally tested for their compliance with the flow continuity principle. Mean flow velocity was inversely related to the ratio of luminal area to regional LV myocardial mass (Equation 2, Fig. 5), which was very similar to the reported by a previous invasive Doppler wire study 16  TAFE versus TAG for detection of DS ≥ 50%. We tested whether TAFE-derived CBF per myocardium (g) is better than simple intra-arterial iodine profile by TAG for discerning obstructive artery from non-obstructive artery. There was no difference among TAGs of DS = 0%, 1-49%, or 50-69%. Only TAG of DS ≥ 70% was significantly lower than that the other TAGs (p < 0.05, all) (Fig. 6A). The optimal cutoff value of   (Fig. 6B).

Discussion
To the best of our knowledge, this study is the first successful demonstration of comprehensive non-invasive coronary physiology assessment. TAFE-derived CBF was validated in both computational simulation model and human perfusion CT study. In the CCTA study, CBF per myocardium (g) consistently decreased according to a gradual increase in stenosis severity, whereas flow velocity was maintained by the compensatory reduction of microvascular resistance and luminal dimension. These non-invasive study results could replicate the results of previous invasive physiology studies 2,16,17 and matched well with the counter-intuitively maintained TIMI 3 flow in a wide range of DS (0-90%). Importantly, all these calculations could be performed from readily and rapidly available parameters sourced from the current conventional CT suite without using any stress agents, pre-specified image acquisition conditions, or heavy computational resources. In addition, CBF per myocardium (g) outperformed TAG for the detection of obstructive coronary artery.    Table 2. TAFE-derived coronary blood flow in perfusion CT and CCTA. * Not all RCA was available (n = 14) because the Z-axis coverage of perfusion imaging was 73 mm. **Hypoplastic vessels (n = 7) were not included. TAFE versus TAG. TAG consistently showed poor correlation with anatomical or physiological stenosis, which has been poorly understood [8][9][10] . Our findings indicated that TAG reflects the basal CBF rather than the coronary pressure gradient 20 . Our results also explain the discrepancies among TAG and DS or FFR, which otherwise comply with the well-known discordance among the basal CBF, DS, and FFR 2,21 .
Clinical implications. The major treatment target of coronary artery disease is ischemia-inducing arterial stenosis. In clinical practice, decisions for treatment are often based only on the anatomical angiographic images. However, physiologic study over the past two decades found that mismatch between anatomical and physiological assessment is not uncommon 7,11,17 . Patients without evidence of myocardial ischemia had favorable outcomes without revascularization, and vice versa. That is, physiology guidance is better strategy than anatomy guidance in the treatment of coronary artery disease [22][23][24] . Hence, invasive physiological assessment with FFR is widely advocated in both American and European clinical practice guidelines for decision of revascularization 25,26 . Despite these well-known clinical benefits, the overall FFR penetration rate is only 6.1% of interventions for intermediate coronary artery stenosis with 40-70% diameter stenosis 27 . This significant underutilization of invasive physiology study suggests that there are still lots of room to expand and to lower the threshold of coronary physiology assessment. FFR is a simplified pressure-based surrogate for coronary physiology. However myocardium thrives on CBF, not on perfusion pressure 21,28 . Although FFR is the acronym of "fractional flow reserve" and represents a deficit in the maximal flow, FFR is limited as a key physiology surrogate in diffuse atherosclerosis, acute coronary syndrome, diabetes, chronic kidney disease, or microvascular angina, all of which show increased basal CBF or marked variation in arterial luminal size 19,[29][30][31][32][33] . In such cases, coronary flow reserve (CFR) and index of microvascular resistance (IMR) have additive or complementary role for prognostic implication 17 Table 3. Coronary blood flow, flow velocity, and microvascular resistance derived from CCTA. p for trend denotes comparison within each column. performing comprehensive invasive physiology study that includes the measurements of FFR, CFR, and IMR for all patients is not practical in clinical medicine considering that only a fraction of patients with coronary artery disease receives clinical benefit from catheterization and revascularization 37,38 . Our methodology enables providing comprehensive physiological data of coronary artery disease, which would expand the utilization and lower the threshold of physiology study and is paramount to good clinical decision-making and improving patient outcome. As additional usage of CCTA-based FFR in clinical practice resulted in less use of invasive procedure and lower healthcare cost compared with conventional practice 38 , CCTA-based comprehensive physiology study may improve the risk stratification of patients, particularly those with diabetes, chronic kidney disease, or diffuse atherosclerosis in which FFR alone may be insufficient for prognostic implication. In addition, non-invasive nature of this methodology enables extraction of physiologic data from preceding CCTA data and large-scaled population-level physiology studies.

Limitations.
Only a single type of CT scanner and a single analyzing workstation was used. As the technical methods and analytical models affect the assessment of positron emission tomography, TAFE coefficient k may be affected by CT scanners and need calibration for each hardware. TAFE may depend on temporal homogeneity, although comparable intracoronary luminal attenuation results between single-beat and multi-beat scans have been recently demonstrated 39 . The spatial and temporal resolution of CCTA causes partial volume effect and limits the accuracy of TAFE in small vessels, which may be overcome by mathematical correction 18,20,40 . Exclusive assumption of laminar flow may cause overestimation of flow velocity or underestimation of microvascular resistance 15 . Invasive physiology study retrieves pressure and flow data from a single distal point, whereas the region of interest in TAFE is not localized to one point. Finally, non-invasive physiology parameters were not directly compared with invasive measurements.