Computed tomographic evaluation of myocardial ischemia

CTP evaluates the first pass of contrast medium through the myocardium at rest and during pharmacological stress. CTP emphasizes the differences in perfusion between normal and ischemic myocardium (Fig. 2) [21]. Myocardial perfusion is a complex process involving the coronary arteries and microvasculature. Coronary artery stenosis decreases the myocardial blood flow and perfusion pressure whereas the microvessels dilate to decrease resistance and maintain resting myocardial perfusion even at 80% luminal stenosis [22]. Autoregulation is limited at rest when the stenosis is severe and during stress even in the earlier stages due to higher myocardial oxygen consumption in these states [22].

Hyperemia is induced by intravenous administration of a vasodilator stress agent, such as adenosine, adenosine triphosphate (ATP), dipyridamole, or regadenoson. Adenosine binds to the A2A adenosine receptor, leading to coronary vasodilation. It is injected at a rate of 0.14 mg/kg/min for 6 min and has a rapid onset of action and a short half-life of 1–5 s. ATP is metabolized to adenosine and acts in the same way as adenosine. It is injected intravenously at a rate of 0.16 mg/kg/min for 5 min and has a half-life of ≤ 20 s. Both these agents are cleared by cellular mechanisms. Dipyridamole acts indirectly by preventing intracellular reuptake and transport of adenosine by endothelial cells and by increasing endogenous adenosine levels. It is injected intravenously at a rate of 0.14 mg/kg/min for 4 min, has a half-life of 30–45 min, and is cleared by liver. Adenosine, ATP, and dipyridamole are non-selective adenosine receptor agonists and cause flushing, headache, lightheadedness, and chest discomfort, and these side effects are naturally treated due to the very short half-life of adenosine or ATP. Regarding dipyridamole, aminophylline (50–250 mg intravenously at least 1 min after the tracer injection) is used if necessary. Regadenoson is a selective A2A agonist that has a lower risk of side effects. It is injected as a bolus of 0.4 mg over 10 s and is followed by a saline flush; it has a half-life of 33–108 s and is cleared by the kidney [23‐25]. Patients should be instructed to avoid consuming caffeine-containing products for ≥ 24 h prior to the scheduled test to prevent the interference with the coronary vasodilatory effects.

Visual assessment is the main evaluation method for static CTP. A perfusion defect is seen as an area with attenuation that is lower than that of remote myocardium in a subendocardial or transmural distribution. The perfusion defect in ischemia could be seen only on the stress CTP images. The perfusion defect in infarction could be seen on both stress and rest images, but the infarction is sometimes missed in the rest images, especially in stress-first protocol. Therefore, LIE-CT is required to identify the infarction [21]. CTP has higher spatial resolution and diagnostic performance for detecting a perfusion abnormality than single-photon emission tomography (SPECT) [54] (Fig. 6, movie in Online Resource 1). The ischemia can be scored semi-quantitatively at a segmental level, similar to SPECT [55]. The American Heart Association recommends that the left ventricle be divided into 17 segments for regional analysis of myocardial perfusion [56]. The scores of the 17 segments can be added to give the summed stress and rest scores, with their difference being the summed difference score [56].

False positives are seen on CTP due to several types of artifact. The most common is beam-hardening artifact, which is caused by preferential attenuation of low-energy photons in a polychromatic X-ray beam. Beam hardening is usually transmural and always occurs in the plane of the X-ray beam adjacent to highly attenuating structures. On CTP, beam-hardening artifact is caused by dense contrast in the left ventricular cavity and descending aorta adjacent to the free wall of the left ventricle and is most commonly seen in the basal inferior and anterior walls of the left ventricle [21, 55]. These artifacts can be minimized by using dedicated beam-hardening correction algorithms or DECT [47, 57]. False-positive perfusion defects may also be seen due to cardiac or respiratory motion or reconstruction artifacts, such as cone beam artifact. Misalignment artifacts can produce image stacks acquired during different heartbeats [21]. Semi-quantitative perfusion measurements have been developed for static CTP and include the transmural perfusion ratio (TPR) and myocardial perfusion reserve index (MPRI) [58]. The TPR and MPRI are calculated as follows: TPR = subendocardial mean attenuation density (AD)/subepicardial mean AD, where MPRI = (AD stress − AD rest)/AD rest. This semi-quantitative assessment has diagnostic performance comparable to that of visual assessment for detection of obstructive CAD [59].

Dynamic CTP can be assessed visually in the same way as static CTP (movie in Online Resource 2). However, dynamic CTP is mainly assessed by a semi-quantitative or fully quantitative technique, because the visual interpretation of CTP image is more difficult than that of MRI due to the lower image contrast [60]. Semi-quantitative parameters are derived from the myocardial time attenuation curve and include upslope, peak enhancement, time to peak, and area under the curve. Upslope has been shown to have the highest diagnostic performance for detection of a perfusion abnormality [61]. Fully quantitative parameters are derived from both arterial and myocardial time attenuation curves, including myocardial blood flow (MBF), myocardial blood volume, and mean transit time. MBF is the most important quantitative parameter and is frequently used for assessment of perfusion whereas the usefulness of myocardial blood volume and mean transit time is not well studied [62] (Fig. 7). Quantitative parameters are useful for assessment of balanced ischemia in triple-vessel disease, which is difficult to assess on SPECT. There are several mathematical methods for quantifying myocardial perfusion in dynamic CTP, including maximum upslope, a compartment model, an extended Toft model, a Patlak plot, a Fermi parametric model, and model-independent deconvolution [35, 63, 64]. However, the optimal quantification method remains controversial and the cut-off value on dynamic CTP imaging is still not standardized. Recently, relative flow reserve has been shown to be a better alternative to absolute MBF [40, 65, 66]. CFR is calculated as a quantitative ratio of stress MBF to rest MBF in stress and rest dynamic CTP protocol [36]. CFR provides the information of not only epi-coronary artery but also microvascular function, and is significantly impaired due to obstructive epicardial CAD or coronary microvascular dysfunction [67]. Obara et al. reported that CFR and stress MBF had sufficient sensitivity, PPV, and NPV to detect obstructive CAD assessed by invasive coronary angiography both in per-patient and in per-vessel analyses as with ¹⁵O-labelled water PET [68, 69]. Nakamori et al. reported that CFR had the additive value in addition to the stress-rest perfusion MRI for detecting reduced FFR in multivessel disease [70]. van de Hoef et al. reported that discordance between CFR and FFR originated from the involvement of the microvascular function, and the risk for major adverse cardiac events associated with FFR/CFR discordance was mainly attributable to stenoses with abnormal CFR [71]. Although CFR is very useful as diagnostic and prognostic predictor, stress and rest dynamic CTP protocol has restriction of use due to the radiation exposure. Therefore, further radiation reduction of dynamic CTP is required to use CFR in clinical routine.

Radiation exposure and suboptimal image contrast are important concerns in CTP imaging. Iterative reconstruction algorithms decrease image noise, allowing use of low-dose CT techniques (e.g., low tube voltage) with maintenance of image quality [72, 73]. There are several iterative reconstruction (IR) algorithms; some are hybrid IR” techniques that are combined with filtered back projection (FBP) to reduce reconstruction time and some are “fully IR” with higher performance [74]. Use of a fully IR technique improves image quality without altering hemodynamic parameters on low-dose dynamic CTP when compared with FBP or hybrid IR [39]. Recently, artificial intelligence was used to improve CT image reconstruction [75]. These techniques have the potential to optimize the quality of low-dose myocardial CTP images with shortening of reconstruction times [76, 77].

A major advantage of CTP is that it allows a perfusion defect to be visualized directly with high temporal and spatial resolution, as would MRI [78]. CTP is useful for assessment of the hemodynamic significance and effective classification of coronary lesions, allowing integrated evaluation of CAD when combined with CTA (Fig. 8) [30, 79]. Moreover, myocardial perfusion can be quantified and the effect of treatment after revascularization can be evaluated [35, 80]. CTP can provide functional information even in patients with heavy calcifications or stenting, which are limitations of CT-FFR [41, 81]. CTP is preferred by patients [82] and is more cost-effective than SPECT [83].

CTP is not widely available because it requires a high level of expertise and multiple resources, including advanced scanners and reconstruction algorithms. Greater radiation exposure, use of a higher contrast dose compared with CTA alone (static CTP alone 1.9–15.7 mSv, dynamic CTP alone 3.8–12.8 mSv, combined CTA and stress CTP protocol 3–16 mSv) [37, 84], and side effects of medications are also issues. Static CTP is limited in evaluation of patients after CABG because of the complexity of myocardial perfusion via the native coronary arteries and bypass grafts.

Multiple studies have established the high diagnostic performance of both static and dynamic CTP for detecting hemodynamically significant stenosis [34, 36, 41, 61, 62, 78, 81, 84‐90] (Table 2). Recent meta-analyses indicated that dynamic CTP has higher sensitivity but lower specificity than static CTP (sensitivity, 85% vs 72–80%; specificity, 90–93% vs 81–83%) [84, 90] (Table 2). Moreover, the CORE320 study (n = 381) demonstrated that static CTP has incremental diagnostic value over CTA for detection of hemodynamically significant coronary lesions, with a specificity of 74% vs 51% for CTA, a PPV of 65% vs 53%, and an area under the receiver-operating characteristic curve (AUC) of 0.87 vs 0.84 [91]. Pontone et al. (n = 147) reported similar results in static CTP (a specificity of 95% vs 76% for CTA and a PPV of 87% vs 61%) [92]. The CRESCENT-II trial indicated that incorporation of dynamic CTP imaging as part of a tiered diagnostic approach could improve the clinical value and efficiency of cardiac CT in the diagnostic work-up of patients with stable CAD and was an effective alternative to standard guideline-directed functional testing [93]. Both static and dynamic CTP has also been reported to have incremental predictive value over clinical risk factors and CTA in assessment for future major adverse cardiac events (MACE) [94, 95]. Dynamic CTP was shown to have higher prognostic value for MACE than CTA and CT-FFR, independent of clinical risk factors [96].

CT-FFR calculates the FFR from coronary CTA data at rest using computational fluid dynamics to generate a mathematical model of coronary flow, pressure, and resistance [97]. This modeling relies on the following four important principles and assumptions: (1) under resting conditions, the total coronary flow is proportional to myocardial mass; (2) resting coronary microvascular resistance is inversely proportional to the size of the epicardial coronary arteries; (3) the dilatory response of the coronary arteries to adenosine during ICA is predictable and can be used to create a computational model of the maximal hyperemic state, which is generally simulated by reducing the microvascular resistance by a factor of 0.21, and although adenosine is not required, administration of nitroglycerine is a pre-requisite for measurement of CT-FFR; and (4) solving the complex three-dimensional Navier–Stokes equation that governs fluid dynamics can compute the flow and pressure across the coronary vascular bed [98].

Several large-scale multicenter trials have shown the high diagnostic performance of CT-FFR by a remote analysis service in selected patients with CAD [98‐100]. In this technology, the CTA data are transferred through a secure network to an off-site location. A patient-specific three-dimensional model of the coronary arteries is created after segmentation of the CTA data. Using the above-mentioned mathematical assumptions, a supercomputer performs complex post-processing to solve the equations governing fluid dynamics and blood flow [97, 101]. Simulated hyperemic blood flow and pressure data are then generated, and the results are sent back to the referring institution within a few hours (Fig. 9). On-site vendor-based platforms are also available in some institutions, including a machine learning-based algorithm [102, 103], four-dimensional CT image tracking (registration) and structural and fluid analysis [104, 105], and patient-specific lumped parameter models [106, 107].

However, the off-site CT-FFR using a remote analysis service is recently received with national reimbursement approval in Japan, but the available facilities are strictly limited by requirements. Meanwhile, the on-site CT-FFR is also available only for clinical research.

CT-FFR is presented as a color-coded map of continuous CT-FFR values computed along each coronary vessel (Fig. 10). These values are both specific for a lesion and for the entire coronary tree. CT-FFR results are interpreted in conjunction with anatomic CTA findings, including vessel size, presence and location of stenosis, suitability for revascularization, and other CT-FFR values. Coronary stenosis with a precipitous drop in CT-FFR across the lesion, particularly if < 0.75, is associated with lesion-specific ischemia. A CT-FFR value > 0.8 distal to a stenosis is rarely associated with ischemia. A CT-FFR value between 0.75 and 0.80 is “gray zone” or borderline [101]. It is important to measure the CT-FFR immediately distal to a stenotic vessel because the CT-FFR value in the most distal vessel segment may not necessarily correlate with the functional significance of the stenosis. A gradual drop in pressure along the length of the vessel without focal stenosis, particularly for borderline values, can be a normal phenomenon or due to a small vessel size, inadequate response to a nitrate, diffuse disease, or a serial lesion [97, 101]. There are ICA data showing that diffuse disease can cause a hemodynamically significant drop in the pressure gradient [108] but the CT data are inadequate. The referring clinician should interpret the findings in view of the anatomy, entire physiological model, and symptoms.

CT-FFR is most valuable in patients with moderate (50–70%) stenosis on CTA. CT-FFR clarifies the hemodynamic significance of stenosis in these patients and aids decision-making [109]. If a patient has moderate coronary artery stenosis on CTA and the CT-FFR value is > 0.8, ICA can be avoided (Fig. 11). However, if the value is < 0.8, ICA can be performed with the intention to treat the lesion (Fig. 12). If CTA shows a non-obstructive lesion (< 50% stenosis), the patient is referred for medical treatment and there is no need to perform CT-FFR or ICA. If CTA shows severe stenosis, the patient is referred for ICA without the need for CT-FFR. In patients with multivessel disease or tandem lesions, CT-FFR helps to identify the lesions needing revascularization [110]. However, CT-FFR may underestimate the contribution of true stenosis in serial stenoses [111]. A noninvasive PCI planning tool has been devised to help determine the contribution of true stenosis in serial and diffuse CAD [111]. A virtual stent can also be placed and the response to revascularization estimated [112].

The accuracy of CT-FFR has been validated in several studies (Table 3) [81, 98‐100, 103, 105‐107, 113‐122], predominantly with off-site technology, which uses invasive FFR as the gold standard. The specificity of CT when CT-FFR is used is better than when CTA alone is used [82% vs 40% in DISCOVER-FLOW (per-vessel), 54% vs 42% in DEFACTO (per-patient), and 79% vs 34% in NXT (per-patient)] [98‐100]. Similar results were seen with on-site vendor-based and machine learning algorithms [81, 103, 105‐107, 113, 116‐119]. One meta-analysis showed a pooled specificity of 76% and an odds ratio of 26.2 for detecting ischemic lesions on the per-patient basis [121], whereas another study showed that CT-FFR had higher specificity than CTA (71% vs 39%) but similar sensitivity (91%) on the per-patient basis [122]. CT-FFR had diagnostic accuracy similar to that of SPECT but had higher sensitivity for predicting FFR-guided revascularization [114]. With good-quality CTA images, CT-FFR was found to have higher diagnostic performance than CTA, SPECT, or PET on a per-vessel basis whereas PET had favorable performance on per-patient and intention-to-diagnose analysis [115].

The studies for CT-FFR are summarized in Table 4. The ability of CT-FFR to evaluate lesion-specific ischemia with high specificity makes it an effective gatekeeper for ICA. In NXT, 68% of false positives were reclassified as true negatives [100]. PLATFORM, which was a clinical utility trial, showed that use of CT-FFR resulted in cancellation of ICA in 61% of patients in whom it had been planned based only on CTA findings without any adverse events at the 90-day follow-up [123]. By using CT-FFR, the overall incidence of non-obstructive disease on ICA decreased to 12% from 73% (a decrease of 61%) [123]. At the 12-month follow-up, adverse events were infrequent (only 5 of 581 followed cases). In the planned invasive stratum, mean costs were 33% lower with CTA and CT-FFR [124]. Data are also emerging on the use of CT-FFR in decision-making and outcomes. The FFR_CT RIPCORD study showed that use of CT-FFR resulted in a change in the management plan (medical vs PCI vs CABG) in 36% of patients. There was a 30% reduction in PCI, an 18% change in the target vessel, and reassignment from optimal medical therapy to PCI in 12% of cases [125]. The SYNTAX III Revolution trial showed that CT-FFR aids decision-making without need for ICA in patients with left main or 3-vessel CAD. A decision-making based on CTA and CT-FFR showed high agreement with the decision derived from ICA [126]. This study also showed that the addition of CT-FFR to CTA alone changed the treatment decision (between PCI and CABG) in 7% of the patients and modified selection of vessels for revascularization in 12% [127]. Using CT-FFR resulted in 12% fewer MACE at 1 year and 30% lower costs with improved quality of life in comparison with ICA and visual guidance [128]. There was no MACE at 1 year in patients in whom ICA was deferred as a consequence of a negative CT-FFR, indicating that this is a safe and feasible test [113, 125]. The 1-year outcomes in the ADVANCE FFR_CT registry study also indicated a low event rate, fewer MACE, and less revascularization in those with negative CT-FFR [129].

CT-FFR shows promise in the evaluation of biomechanical forces on atherosclerotic plaques, which play a role in their development and progression [130, 131]. CT-FFR can also aid in the prediction of acute coronary syndrome with potentially superior accuracy (area under the curve, 0.725) on CTA [132].

CT-FFR relies on high-quality CTA images. Therefore, it is critical to adhere to a rigorous guideline-driven protocol [133] with maximal coronary vasodilation and without image noise, motion, or misalignment artifacts [101, 134]. Inadequate contrast opacification and calcium blooming can also negatively affect CT-FFR evaluation [135]. However, recent studies have shown that the performance of CT-FFR in patients with high coronary calcium score is superior to that of CTA alone [130, 136]. The rates of rejection on CT-FFR analysis in the ADVANCE registry and in a large clinical cohort were 2.9% and 8.4%, respectively [137]. CT-FFR is currently not suitable for patients with stenting or CABG. CT-FFR is also not effective in patients with unstable angina and performs modestly in detecting ischemia in non-culprit lesions in patients with STEMI, probably because of the vessel volume is smaller in these patients than in those with stable angina, which confounds the assumption that size is related to resistance [138]. As of now, CT-FFR is not ideal for use in the emergency setting because of the above-mentioned inadequate assumptions. A recent study showed that although CT-FFR has good accuracy overall (81.0%), it has poor accuracy (46.1%) in the borderline CT-FFR range (0.7–0.8) [139]. Therefore, the results of CT-FFR should be interpreted cautiously in the context of patient-specific risk factors. Furthermore, there are data that show CT-FFR to be abnormal in up to 16.6% of patients with insignificant stenosis (< 50%) and conversely that it can be normal in 50% of patients with moderate stenosis [140, 141]. Recent studies have cast doubt on the reproducibility of CT-FFR [138]. For example, there is not a perfect match between the CT-FFR and invasive FFR values [139]. This could be because the pressure sensor on invasive FFR is not at the exact location where CT-FFR was measured [101] or the dose of nitroglycerin before CTA was inadequate or nitroglycerin was not administered at all [101].