PET relies on the simultaneous detection of two photons, emitted from the decay of radionuclide tracers. In more detail, positrons are emitted during the distribution of these tracers in the patient’s body and collide with an electron. Consequently, a positron annihilates, which results in the emission of two photons in opposite directions. Since the average range traveled by positrons is small, in the order of mm, the decay can be considered to have occurred along the straight line described by the two annihilation photons. A PET scanner contains several rings of detectors, made of a scintillating material, which convert the energy of the annihilation photons proportionally into an electrical signal. Two photons are considered to have been emitted simultaneously when they are detected within the narrow coincidence-timing window of the scanner, around 6–12 ns. Accordingly, when two photons are detected simultaneously, a decay event should have occurred somewhere along the line between two detectors. By detection of these annihilation photon pairs, the distribution of the positron-emitting nuclides in the patients part positioned within the field-of-view of the PET scanner can be reconstructed.

As compared with single-photon emission computed tomography (SPECT), PET has several advantages. One of the main benefits is the superior image quality of PET over SPECT. This improvement is due to more favorable tracer characteristics, improved count statistics, as well as the routine and more accurate application of photon attenuation correction (AC). Although this correction technique is increasingly available for SPECT imaging with possible benefit in terms of diagnostic accuracy, the downside is that it can also induce artifacts [1]. Furthermore, smaller and more subtle perfusion defects can be detected due to higher spatial resolution of PET (typically 4–7 mm) as compared to SPECT (typically 12–15 mm). Next to spatial resolution, also temporal resolution is in favor of PET, which allows for absolute quantification of perfusion by tracking the dynamic tracer activities of arterial blood and myocardium through time. Although it has been attempted for SPECT [2], PET is an established tool to provide clinically relevant quantitative levels of perfusion and flow reserve next to qualitative myocardial perfusion images [3‐6]. Other advantages include a lower radiation burden and acquiring both rest and stress images within a single scanning session due to the short physical half-life of the PET perfusion tracers. The main limitation for PET is the need for an on-site cyclotron or generator with the current tracer agents as will be discussed in more detail. An overview of PET and SPECT imaging characteristics are provided in Table 1.

Of several available PET tracers, ⁸²Rb, ¹³NH₃, and H₂ ¹⁵O are the most commonly used for the assessment of myocardial perfusion [3]. Additionally, ¹⁸F-flurpiridaz is an emerging perfusion tracer which holds great clinical potential but is not yet available for clinical use and is currently being tested in phase 3 trials [7‐9]. Tracer specific characteristics, including pros and cons, will be described below and summarized in Table 2. It’s important to realize that none of the perfusion tracers excels on all of these features. Therefore the choice of tracer is multifactorial and depending on practical considerations, as well as the aim of the PET imaging program.

Among clinically used perfusion tracers, H₂ ¹⁵O features fundamentally different properties compared to ⁸²Rb, ¹³NH₃, and ¹⁸F-flurpiridaz [3, 6, 10]. Namely, ⁸²Rb is a potassium analog and is taken up by myocardial cells via the Na/K ATP transporter in a rapid and active manner [11]. While ¹³NH₃ is incorporated into the glutamine pool through active transport and passive diffusion processes [12]. ¹⁸F-flurpiridaz is derived from pyridazinone and binds avidly to mitochondrial complex-1 [13]. In other words, these three tracers are transported across the cell membrane and effectively become metabolically trapped while they are cleared from the intravascular compartment (arterial blood pool). Consequently, ‘late’ static uptake images of these tracers account for high tissue-to-background ratios and result in excellent qualitative grading of relative perfusion distribution. The combination of these images with ECG-gating permits the assessment of left ventricular (LV) volumes and function as well as regional wall motion [14].

In contrast, H₂ ¹⁵O diffuses freely across myocyte membranes, is metabolically inert and thereby promptly reaching equilibrium between blood and tissue without accumulation in the myocardium. As a consequence, radiotracer distribution images of H₂ ¹⁵O are of poor image quality and provide little diagnostic value. The lack of diagnostic images has long prohibited the use of H₂ ¹⁵O in clinical practice and nearly all studies on qualitative imaging for CAD have been conducted with ⁸²Rb or ¹³NH₃ [15]. In recent years, however, digital subtraction techniques and parametric imaging by automated software packages now generate qualitative gradable images that display perfusion at a voxel level, based on the tracer kinetic model for each voxel [16, 17]. These images are distinctly different from actual tracer uptake images as they represent a graphical illustration of quantitative flow values. These developments have enabled H₂ ¹⁵O to be utilized in clinical practice [18, 19].

Image quality is additionally determined by the positron range in tissue. High-energy positrons penetrate deeper into tissue before annihilation occurs and demonstrate decreased spatial resolution compared to low-energy positrons. Therefore, image resolution gradually increases from ⁸²Rb, H₂ ¹⁵O, ¹³NH₃, to ¹⁸F-flurpiridaz, respectively, according to their energetic state (Fig. 1) [4]. Moreover, the physical half-life of the radioactive compounds determines the potential acquisition duration and therefore count-statistics. The short physical half-life of ⁸²Rb and H₂ ¹⁵O allows a timeframe of only a few minutes of acquisition before the tracer is decayed to background levels, whereas ¹³NH₃ and ¹⁸F-flurpiridaz acquisitions can be continued till satisfactory counts-statistics are obtained, which enhances image quality. These factors result in the highest image quality of ¹⁸F-flurpiridaz given its long half-life and low positron range as opposed to relative poor image quality of ⁸²Rb with its ultra-short half-life and high positron range.

Next to relative uptake images, PET enables the assessment of absolute levels of tracer concentration. Using a dynamic acquisition (i.e. multiple frames initiated upon administration of the tracer) time-activity curves can be generated of tracer flux for arterial blood and myocardium. Automated software then computes myocardial blood flow (MBF) in absolute terms (in units of mL·min⁻¹·g⁻¹) and calculates coronary flow reserve (CFR) [16, 17]. An ideal tracer for these measurements is characterized by accumulation in/or clearance from myocardium proportionally linear to perfusion, irrespective of flow rate or metabolic state [20]. H₂ ¹⁵O is the only tracer that meets these criteria and is considered the gold standard for quantification of MBF [21]. The other aforementioned tracers have the property that myocardial extraction from arterial blood is incomplete and curvilinear with increasing flow rates, frequently referred to as the “roll-off” phenomenon (Fig. 2) [22]. PET derived MBF measurements are therefore underestimated with increasing actual flow. Correction models based on animal experiments can be employed yet induce noise, particularly when large correction factors are required with severely blunted extraction at high perfusion levels or with tracers characterized by a lower extraction fraction. Nonetheless, each of these tracers has been tested in animal experiments against microsphere-quantified perfusion, the invasive reference standard. H₂ ¹⁵O and ¹³NH₃ in particular have been well validated and display close agreement with microsphere flow and demonstrate low test–retest variability (10–15%) [10, 21, 23‐25]. In recent years, automated software packages have been developed and improved, applying these validated models. Postprocessing is now in the order of minutes, and these packages display high reproducibility [16, 26, 27]. Quantification of ⁸²Rb is less reliable as this tracer harbors intrinsic limitations (ultrashort physical half-life, long positron range, and low extraction fraction). Nonetheless, its extraction fraction is still superior to Technetium-99m (^99mTc)-labeled SPECT (Fig. 2) and recent studies have shown MBF measurements of ⁸²Rb to be feasible [28]. Limited data are available concerning the quantification of ¹⁸F-flurpiridaz, but its characteristics and kinetics (very high extraction fraction and short positron range) have the potential for highly reliable perfusion measurements [8, 20, 29].

Of interest, recent developments enabled the estimation of quantitative MBF using alternative noninvasive imaging techniques, such as cardiovascular magnetic resonance imaging (CMR) and CT, with the use of contrast media [31, 32]. The low extraction fraction of these iodine and gadolinium based contrast agents (Fig. 2), however, necessitates the use of extensive corrections and limits the accuracy of MBF measurements.

A fundamental concern for widespread clinical application of cardiac PET perfusion imaging is the necessity to produce the utilized tracers onsite. In this regard, the currently available tracers H₂ ¹⁵O and ¹³NH₃ require a cyclotron in close proximity of the scanning facilities. Additionally, while H₂ ¹⁵O is mainly used in European and Asian nuclear imaging labs at this time, FDA hasn’t approved this tracer for clinical use. ⁸²Rb, however, is produced by a ⁸²Sr/⁸²Rb generator obviating the need for a cyclotron and is therefore more convenient to implement in clinical practice. The downside of this approach is that the generator needs to be replenished every 28 days at relatively high costs ($20,000). In order to make such a program cost-effective, a high volume patient throughput is needed. These issues may soon be overcome by the emerging fluorine-labeled tracers such as ¹⁸F-flurpiridaz [8]. Because its longer physical half-life of 110 min allows for off-site production, this tracer has great potential for widespread implementation. ¹⁸F-labeled perfusion tracers also benefit from the fact that they can be used in physical exercise protocols whereby the radioisotope is administered during maximal exertion. ⁸²Rb, H₂ ¹⁵O, and ¹³NH₃ require injection while the patient is lying inside the scanner, as tracer decay is too rapid to transport the patient from the treadmill or stationary bike to the scanner. These tracers can therefore only be utilized in conjunction with pharmacological stressor agents.

Similar to SPECT, in clinical practice PET perfusion images are most commonly graded visually and in a qualitative manner. Relative radiotracer distribution is assessed during both rest and stress (or hyperemic) conditions. Myocardial perfusion defects are usually graded by their extent, severity, and location. Current guidelines recommend a semiquantitative analysis using a segmental 5 point scale system (normal = 0, mild defect = 1, moderate defect = 2, severe defect = 3, and absent uptake = 4) on a 17 segment model of the left ventricle [33, 34]. These scores can be summed for rest (SRS) and stress (SSS) with a subsequent summed difference score (SDS) in order to identify reversibility [35]. Fixed defects are compatible with myocardial scarring or hibernating myocardium, whereas reversibility of stress induced hypoperfusion is compatible with ischemia.

Next to qualitative and semiquantitative grading, PET also allows for absolute quantification of perfusion. Several available automatic software packages routinely provide these MBF values per myocardial territory. Derived MBF values can then be compared with normalcy ranges of flow. Normal databases, however, display a broad base of hyperemic MBF between 2 and 5 mL min⁻¹ g⁻¹, which is attributable to variability in minimal microvascular resistance and is dependent on age, sex, and traditional cardiovascular risk factors [36‐39]. Currently, still limited data are available with regard to an optimal threshold to distinguish pathological from normal hyperemic MBF and myocardial flow reserve [5]. In addition, thresholds for PET derived MBF values are not interchangeable for different radiotracers. Although in general, a myocardial flow reserve below two is considered abnormal whereas beyond 2.5 is deemed normal, with an ambiguous transition zone between 2.0 and 2.5 [6]. These values were confirmed by a recent multicenter study presenting an optimal threshold of 2.30 mL min⁻¹ g⁻¹ for hyperemic MBF and 2.50 mL min⁻¹ g⁻¹ for myocardial flow reserve when compared with invasive fractional flow reserve (FFR) measurements [40]. It can be questioned, however, whether single thresholds are reasonable. Alternatively, MBF values might be interpreted on a continuous scale for diagnostic and prognostic purposes as well as subsequent clinical decision making [5]. Therefore ongoing studies are targeted to further define the normal limits of (hyperemic) perfusion, especially for different subgroups such as revascularized patients as well as patients with diabetes and cardiomyopathies [41]. Another important issue is that myocardial perfusion imaging reflects the composite of the epicardial as well as the microvascular bed. This means that diminished flow values may originate from either epicardial or microvascular disease, or both.

The majority of studies exploring the diagnostic accuracy of PET perfusion imaging for the detection of CAD, have been conducted with static uptake images of ⁸²Rb and ¹³NH₃. Compared with SPECT, perfusion imaging using PET consistently yields the highest diagnostic accuracy [42‐44]. Sensitivity and specificity for PET in these meta-analyses ranged from 84 to 93% and 81 to 88%, respectively. It must be acknowledged, however, that most of these studies were compared with invasive coronary angiography without FFR and therefore lack an appropriate reference standard.

Diagnostic accuracy testing has been less extensive for quantitative perfusion imaging. Increasing data, however, show the superiority of quantitative assessment over static uptake image grading [45‐49]. Typical groups of patients which could benefit the most from quantitative assessment include patients with multivessel disease (i.e. balanced ischemia), early stage blood flow impairment, and microvascular disease [6, 47, 50]. Especially multivessel disease frequently results in false negative interpretation of relative radiotracer uptake, because the myocardial region with the highest uptake is considered the normal reference region. Absolute blood flow quantification would then reveal that this region is abnormally perfused as well. This is illustrated with an example in Fig. 3. Apart from this, the combination with ECG gated derived information such as LV function and transient ischemic dilatation (TID) seems to increase diagnostic accuracy of qualitative uptake images [51, 52].

Another interesting finding from recent studies is that hyperemic MBF quantification outperforms CFR to diagnose obstructive CAD, highlighting the potential of stress only protocols [40, 53, 54]. The largest of these studies, involving 330 patients, reported a sensitivity, specificity, and accuracy of 89, 84, and 86% against 86, 72, and 78% for hyperemic MBF and CFR, respectively [40] (Fig. 4).

Studies determining the prognostic significance of SPECT contain larger databases, nonetheless parallel the value of PET [55]. The extent and severity of PET derived perfusion defects have also been demonstrated to hold strong prognostic information beyond traditional cardiovascular risk factors [56, 57]. In addition, the quantitative nature of PET has the potential to further increase prognostic significance. Due to greater availability, especially quantitative ⁸²Rb and ¹³NH₃ PET have shown incremental value for predicting adverse cardiac events over traditional relative perfusion imaging grading [4, 58‐60]. Murthy et al. [4] revealed a significant association between quantitative coronary flow reserve (CFR) measurements and cardiac mortally even after adjustment for traditional risk factors and visual perfusion imaging grading in nearly three thousand patients undergoing ⁸²Rb PET imaging. CFR measurements also induced correct reclassification of estimated risk categories in 35% of patients with a previously intermediate risk on cardiac death. Of particular interest is reclassification of perfusion images with visually homogenous tracer distribution caused by diffusely blunted hyperemic perfusion. Several studies have revealed that this subset of patients is at increased risk for future cardiac events [58, 59, 61, 62] Additionally, reduced flow values predict higher risk of cardiac events, even without obstructive CAD. Microvascular dysfunction is thought to play an important pathophysiological role in these patients [63, 64].

Current PET scanners are virtually always equipped with a CT component. These hybrid PET/CT devices are now available up to 128-slice CT and offer near simultaneous assessment of comprehensive anatomical and functional information within a single scanning session, which can be as short as 30 min. An example is shown in Fig. 5.

Coronary computed tomography angiography (CCTA) is an established tool for the non-invasive detection of coronary atherosclerotic stenosis. A multitude of studies have shown a high diagnostic performance of CCTA for the identification of coronary artery stenosis [65‐67]. In particular, its sensitivity and negative predictive value are consistently demonstrated to be near perfect, approximating 100%. CCTA is therefore currently the ultimate modality to exclude CAD in patients with a low to intermediate pre-test likelihood of disease. Furthermore, CCTA enables noninvasive evaluation of plaque morphology. Thereby detecting very early stages of CAD as well as plaques that might be vulnerable for rupture [68]. The specificity is, however, hampered as stenosis severity is often overestimated [69]. Another downside of CCTA concerns its limited ability to predict hemodynamic consequences of atherosclerotic stenosis.

On the contrary, myocardial perfusion imaging using either PET or SPECT, is particularly useful to assess hemodynamic significances and thus document myocardial ischemia. As mentioned before, without the knowledge of coronary anatomy and (the location of) stenosis, the results of perfusion imaging should be interpreted carefully.

In summary of the above, either a solely anatomical or functional approach in the evaluation of CAD has its limitations. Therefore, a hybrid assessment could provide complementary rather than overlapping information. The limited number of studies on the diagnostic value of PET/CT seem to confirm the theoretical enhanced accuracy as compared with either modality alone [18, 19, 70, 71]. Other studies revealed analogous improvement of diagnostic performance when fusing SPECT and CCTA [72, 73]. It is shown that especially the moderate specificity of CCTA benefits from the use of hybrid imaging and results in a more judicious referral pattern for invasive coronary angiography [74‐76].

In the absence of prognostic data on hybrid PET/CT, the results from hybrid SPECT/CT studies indicate an enhanced risk stratification compared to standalone modalities [75, 77, 78]. This holds particularly true when either perfusion or angiographic imaging exhibit equivocal results. Kim et al. [77] demonstrated incremental prognostic value of sequential SPECT and CCTA in 1295 patients with suspected CAD. However, there was no significant additive value in the case of either stenosis ≥90% on CCTA or SSS ≥ 4 on SPECT. Figure 5 illustrates an example of both standalone modalities as well as hybrid imaging in comparison with invasive coronary angiography.

Interestingly, CT and CMR imaging provide alternative approaches with the possibility of assessing both anatomy and perfusion using a single imaging modality. For CT, the assessment of coronary anatomy can be combined with myocardial perfusion. But the acquisition of such a dynamic first pass sequence comes at the cost of high patient radiation burden, next to the aforementioned unfavorable contrast agent characteristics for CT. A recent multicenter trial, however, showed that CT perfusion using a 320-slice CT scanner improved diagnostic accuracy over CCTA alone [79]. CMR does not have the issue with ionizing radiation and can be combined with e.g. the evaluation of LV and valvular function. In a recent trial, the diagnostic accuracy of CMR in the detection of CAD was found to be superior to SPECT [31]. Still, also CMR perfusion faces multiple technical issues such as imaging artifacts, incomplete coverage of the LV and, as mentioned before, the gadolinium contrast agent impedes accurate MBF measurements with the corresponding disadvantages as compared to PET. Furthermore, not every patient is eligible to undergo CMR because of claustrophobia and contraindications such as pacemakers and implantable cardiac defibrillators.