This retrospective study was approved by the local ethics committee, and the requirement for informed consent to participate this study was waived. The records of 55 patients who underwent FFR evaluation during coronary angiography and SPECT myocardial perfusion imaging because coronary artery disease was suspected by CT angiography from April 2014 to November 2018 were retrospectively examined. The following patients were excluded: interval between SPECT and CT > 3 months (n = 5), interval between catheter exam and CT > 3 months (n = 4), severe motion artifact at coronary CT angiography (n = 1), FFR only performed in a vessel with suspected in-stent restenosis (n = 1), and caffeine intake before coronary angiography (n = 1). Thus, the final study group included 43 patients. Of the assessable 129 vessels, the following two vessels were excluded: FFR not performed despite intermediate stenosis and hypoplastic right coronary artery due to left dominance. The pretest probability of coronary artery disease was estimated by a previously described method [4] .

All patients underwent CT angiography with a 64-row CT (Brilliance 64; Philips, Eindhoven, Netherlands, or Somatom Definition AS+; Siemens Healthineers, Forchheim, Germany).

Patients with heart rate > 65 beats/min at the outpatient department were told to take an oral β-blocker (20 mg of metoprolol) 1 h prior to CT angiography. If the heart rate was over 65 beats/min on site, a maximum dose of 12.5 mg of landiolol (Corebeta; Ono Pharmaceutical, Tokyo, Japan) was given intravenously [5] . All patients received 0.3 mg sublingual nitroglycerin (Nitropen; Nippon Kayaku, Tokyo, Japan) before imaging.

The scanning parameters for Brilliance 64 were as follows. Calcium scoring scan was performed by a prospective electrocardiogram-gated axial scan with 120 kVp, 196 mA and 2.5 mm collimation. Images were acquired at 40% or 70% of the R-R interval depending on the heart rate and were reconstructed with a slice thickness of 2.5 mm and increment of 2.5 mm. Coronary CT angiography was performed by electrocardiogram-gated helical scan with dose modulation technique and the following parameters: Detector configuration, 64 × 0.625 mm; tube potential, 120 kVp; tube current-time product, 800 - 1050 mAs, depending on the body weight; gantry rotation time, 420 ms; and helical pitch, 0.2. The reconstructed slice thickness was 0.67 mm, and the increment was 0.33 mm. Images were reconstructed using a cardiac sharp kernel. Iterative reconstruction technique was not available with this scanner.

The scanning parameters for Definition AS+ were as follows. Calcium scoring scan was performed by a prospective electrocardiogram-gated axial scan with 120 kVp and a reference mAs of 40 mAs. Coronary CT angiography was performed by prospective electrocardiogram-gated axial scan. The tube potential and the reference mAs were set as 120 kVp and 250 mAs. The acquisition window initiated at 60% - 75% of the R-R interval when the heart rate was below 65 beats/min, otherwise initiated at 30% - 75%. The reconstructed slice thickness was 0.75 mm and the increment was 0.4 mm using a cardiac kernel (I36f) with sinogram-affirmed iterative reconstruction strength 2.

For processing, images were transferred to a workstation (Synapse Vincent Ver 5.1; Fujifilm Medical, Tokyo, Japan).

Calcified lesions with a minimum area of three pixels and a minimum CT number of 130 HU were scored by using the algorithm developed by Agatston et al. [6] . Lesion scores from the left main, left anterior descending, left circumflex, and right coronary arteries were summed to determine a total calcium score. Calcium score was not calculated in patients with stents.

When an intermediate stenosis (40% - 70%) was present, EEL was calculated by a previously described method [3] . Briefly, the energy loss across a coronary stenosis was calculated using the following parameters: Lesion length (LL), diameter stenosis (DS), minimal lumen area (MLA), and the myocardial volume of the stenosis-related territory. Lesion length, diameter stenosis, and minimal lumen area were automatically obtained using the stenosis analysis software of the workstation (Figure 1(a)). The left ventricular volume of the stenosis-related territory was estimated using the Voronoi method (Figure 1(b)). The coronary

flow at maximal hyperemia was estimated at 4.0 ml∙min⁻¹∙g⁻¹. Using a simplified Bernoulli formula, pressure loss across a stenosis can be estimated as follows [7] .

Δ P = Q 8 π μ L L M L A 2 + Q 2 ρ 2 ( 1 − ( 1 − D S / 100 ) 2 ) 2 M L A 2

where ∆P is the pressure drop, Q is the blood flow, μ is the blood viscosity (0.004 kg∙m⁻¹∙s⁻¹) and ρ is the blood density (1000 kg∙m⁻³). The first term accounts energy loss due to viscous friction between laminar layers of fluid and the second term reflects energy loss when normal arterial flow is transformed first to high-velocity flow in the stenosis and then to the turbulent nonlaminar distal flow eddies at the exit from the stenosis. We defined EEL as log ∆P and determined EEL > 1.17 as ischemic [3] .

FFR was performed using a pressure wire (PressureWire Cetrus; St Jude Medical Systems, St. Paul, MN) when the stenosis was between 40% - 70% at invasive coronary angiography. The pressure wire was positioned distal to the stenosis of interest under intravenous infusion of adenosine triphosphate (Adetphos; Kowa Company, Tokyo, Japan) at 140 μg∙kg⁻¹∙min⁻¹ for 3 min. FFR was derived as the ratio of the mean coronary pressure distal to the stenosis over the mean aortic pressure at maximal hyperemia. An FFR value of ≤0.80 was determined as hemodynamically significant. The cardiologists performing angiography had total access to the coronary CT data. However, they were not aware of EEL because this value was calculated for research use and not mentioned in the radiology report.

SPECT was performed using ^99mTc-tetrofosmin (Myoview; Nihon Medi-Physics, Tokyo, Japan) with 296 MBq used for stress and 740 MBq used for rest studies. Pharmacologic stress was performed using intravenous infusion of adenosine (Adenoscan; Daiichi Sankyo Healthcare, Tokyo, Japan) at 120 μg∙kg⁻¹∙min⁻¹ for 6 min. Electrocardiographic gating on a dual-headed SPECT system was performed with 8 frames per cardiac cycle (Infinia Hawkeye 4; GE Healthcare, Milwaukee, WI). Two independent observers visually interpreted the images and joint consensus reading was performed when discrepancy was present. Stress-induced ischemia was determined when a perfusion defect at stress was accompanied with a fill-in at rest (Figure 1(c)).

Continuous variables were shown as mean ± standard deviation and categorical variables as number (%) unless otherwise described. The Student t-test was used to compare continuous variables. The Chi square (χ²) test and the Fisher exact test were used to compare categorical and skewed variables. Interobserver agreement for diagnosis of perfusion defect at SPECT was calculated using the Cohen κ statistic [8] , which was interpreted as poor (κ < 0.20), fair (κ = 0.21 - 0.40), moderate (κ = 0.41 - 0.60), good (κ = 0.61 - 0.80), very good (κ = 0.81 - 0.90), or excellent (κ ≥ 0.91). The McNemar test was used to assess the difference in the diagnostic accuracy. Receiver operating characteristics (ROC) curve analysis was performed to compare the diagnostic performance of EEL and SPECT to predict ischemia-causing stenosis by FFR. The difference in area under the curve (AUC) was assessed by the DeLong method [9] using R (The R Foundation for Statistical Computing). The remaining statistical analyses were performed using JMP software (version 12.2.0; SAS, Cary, NC). A p-value < 0.05 was deemed to indicate significance.

The majority of patients (74%) were male and the coronary risk was moderate (Table 1). The median (interquartile range) calcium score was high with a value

Numbers are reported as average ± standard deviation or N (%), unless otherwise described. *The risk score was determined by the criteria by Tomizawa et al. [4] . Median (interquartile range). An FFR value ≤ 0.80 or diameter stenosis ≥ 70% was determined hemodynamically significant. FFR, fractional flow reserve; PCI, percutaneous coronary intervention.

of 468.7 (142.5 - 848.7). Eighteen patients (42%) had at least one vessel with ischemia proven by FFR (≤0.80) or severe stenosis (diameter stenosis > 70%). FFR was measured in 49 vessels and the average value was 0.83 ± 0.10. The ischemic vessel was most frequent at the left anterior descending (14 vessels (33%)), followed by the left circumflex (9 vessels (21%)), and the right coronary artery (3 vessels (7%)).

CT angiography showed severe stenosis (diameter stenosis > 70%) in 8 vessels and the calculated EEL was significant (>1.17) in 21 vessels. Hence, 29 (23%) out of 127 assessable vessels were determined as ischemic by CT. A total of 20 vessels (16%) were determined as ischemic by SPECT. The inter-observer agreement was good (κ = 0.67).

The AUC of the ROC curve of EEL was significantly higher than that of SPECT to predict ischemia-causing stenosis (0.86 (95% confidence interval (CI), 0.78 - 0.95) vs 0.67 (95% CI, 0.57 - 0.77), p < 0.005) (Figure 2(a)). The sensitivity (81% vs 42%, p = 0.009) and negative predictive value (95% vs 86%, p = 0.04) of EEL was significantly higher than SPECT (Figure 3), while the specificity (92% vs 91%, p = 1.0) and positive predictive value (72% vs 55%, p = 0.24) did not show significant difference (Table 2). The diagnostic accuracy of EEL was significantly higher than SPECT to diagnose ischemia-causing stenosis (90% vs 81%, p = 0.04). The accuracy of EEL and SPECT was similar for the left anterior descending (74% vs 72%, p = 0.81), while it was marginally higher in the right coronary artery (98% vs 85%, p = 0.06) and significantly higher in the left circumflex artery (98% vs 86%, p = 0.03) (Table 2).

Numbers in parenthesis represent 95% confidence interval. * Statistically significant, p < 0.05. EEL, estimated energy loss; LAD, left anterior descending; LCX, left circumflex; FFR, fractional flow reserve; RCA, right coronary artery; SPECT, single photon emission computed tomography.

There were 5 false-negatives for EEL (Figure 4) with an average value of 0.93. CT underestimated the stenosis severity in 1 vessel and the average FFR of the remaining 4 vessels were 0.77 (range, 0.71 - 0.80). Conversely, EEL resulted as false-positive in 8 vessels and the average FFR was 0.86 (range, 0.82 - 0.90). The average MLA and EEL were 1.04 mm² (range, 0.73 mm² - 1.76 mm²) and 1.47 (range, 1.21 - 2.02), respectively.

A total of 21 patients were determined to have ischemia by EEL while 15 patients had ischemia by SPECT. The AUC of the ROC curve significantly improved from 0.58 (95% CI, 0.43 - 0.73) to 0.84 (95% CI, 0.74 - 0.95) (p < 0.005, Figure 2(b)). The sensitivity (89% vs 44%, p = 0.01) and negative predictive value (91% vs 64%, p = 0.04) of EEL was higher than SPECT (Table 3). The diagnostic accuracy of EEL was significantly higher than SPECT to diagnose ischemic patients (84% vs 61%, p = 0.01).

Numbers in parenthesis represent 95% confidence interval. EEL, estimated energy loss; FFR, fractional flow reserve; SPECT, single photon emission computed tomography. *Statistically significant, p < 0.05.

There were 2 vessels which SPECT correctly detected ischemia but missed by EEL. The lesion length was very short (3.2 mm) in one lesion and the minimal lumen area was large (3.02 mm²) (Figure 4), in the other lesion. Conversely, EEL overdiagnosed 7 vessels which SPECT correctly ruled out ischemia. The minimal lumen area was small (<1 mm²) or either lesion length was long (>15 mm).

SPECT misdiagnosed 11 vessels as non-ischemic which were ischemic by EEL and FFR. Five of these vessels were from multiple-vessel disease patients. The average FFR of the remaining vessels was 0.77 (range, 0.70 - 0.80). Overdiagnosis of SPECT occurred in 8 vessels which were actually non-ischemic by FFR and correctly diagnosed by EEL. Five of these vessels were in the RCA territory, followed by LAD (2 vessels) and LCX (1 vessel).