Functional cardiac CT–Going beyond Anatomical Evaluation of Coronary Artery Disease with Cine CT, CT-FFR, CT Perfusion and Machine Learning

Abstract

The aim of this review is to provide an overview of different functional cardiac CT techniques which can be used to supplement assessment of the coronary arteries to establish the significance of coronary artery stenoses. We focus on cine-CT, CT-FFR, CT-myocardial perfusion and how developments in machine learning can supplement these techniques.

Introduction

Cardiac CT is a highly accurate test to detect the presence of coronary atherosclerosis and to precisely quantify coronary plaque burden. 30 years ago, Agatston et al published the now eponymous and famous method to quantify the presence and amount of coronary calcium,¹ and many studies have conclusively shown the strong relationship between the presence of coronary calcium and the risk for future coronary events.^2,3 Since the early years of this millennium, aided by the development of multidetector row CT scanners, researchers have widened the focus to also visualize the coronary artery lumen and non-calcified coronary plaque using various coronary CT angiography (CCTA) techniques.⁴ In the past decade, CCTA has further matured and state of the art scanners with which the heart and coronary tree can be precisely depicted in a single short breath hold are now available from all major hardware vendors.

At the same time, it has also become clear that the presence of coronary plaque, even when there is substantial luminal narrowing, does not necessarily imply the presence of a significant pressure gradient or flow limitation⁵ and by extension, the need for invasive intervention. Even in patients with proven moderate-to-severe ischemia and stable chest pain, the indication for invasive intervention has recently been questioned.⁶ The inability of CCTA to precisely judge the hemodynamic significance of coronary artery stenoses using the degree of luminal narrowing alone has sparked the search for alternative methods to improve this assessment. Approaches to bridge this gap range from taking into account the type of coronary plaque to examining myocardial contrast enhancement using the principles of tracer kinetics, and more recently, assessment of cardiac motion and application of advanced computational fluid dynamics modeling as well as machine learning (ML) techniques. Below we summarize these efforts and review their clinical potential.

Functional CT – overview of different techniques

In this review, we will consider three different functional cardiac CT techniques: 1) cine CT to assess cardiac motion; 2) estimation of pressure gradients across coronary stenoses using computational fluid dynamics or other methods; and 3) CT myocardial perfusion. We will not focus on coronary plaque assessment, as this has been covered elsewhere in this special issue by Bing et al.⁷

Cine CT

Since its introduction, retrospectively ECG-gated CT has allowed for evaluation of the heart over the entire cardiac cycle. This multiphasic acquisition technique provides dynamic views (i.e., cine-CT or 4D-CT) for functional analyses during both systole and diastole. In patients with chest pain, retrospectively gated CCTA can provide additional information regarding cardiac function without the use of additional contrast medium, radiation dose or additional tests. CT may also provide functional information in symptomatic patients with (relative) contraindication for cardiac MRI (CMR). A disadvantage of multiphasic CT is the radiation dose, which can range between 6 and 15 mSv depending on the CT acquisition protocol (e.g., use of dose modulation or low tube voltage).⁸

Left ventricular ejection fraction (LVEF) is the most universal tool to evaluate left ventricular function and has been used as diagnostic and prognostic marker for treatment decision-making. CMR provides high spatial and temporal resolution and is the gold standard for LVEF assessment. Nevertheless, over the last years, multiple studies have proven feasibility and accuracy of both CT-derived LVEF^9,10 and right ventricular ejection fraction (RVEF),^9–12 in (acceptable) accordance with CMR results.

Assessment of regional LV function by CT in acute chest pain showed increased sensitivity and improved detection of acute coronary syndrome compared to CCTA alone in a subanalysis of the ROMICAT trial.¹³ Cine CT provides information on wall motion, as reconstructed cine-loop images can visualize myocardial contraction and relaxation throughout the cardiac cycle. In patients with ischemic heart disease, CT can be used as alternative tool to assess global and regional wall motion abnormalities.^14,15 Likewise, the combination of CCTA and functional information may be helpful in patients without chest pain but increased risk for coronary artery disease (CAD) and LV dysfunction, such as in oncologic patients with suspected cardiotoxicity or radiation-induced CAD.^16–18 With further improvement of CT acquisition techniques, ventricular diastolic dysfunction and atrial function may potentially be depicted as well.^19–21

Cine-CT can provide information on myocardial mass, wall thickness and wall motion, by delineation and tracking of the endocardial and epicardial boundaries throughout the cardiac cycle.²² Furthermore, myocardial strain has emerged as valuable parameter for assessment of global LV function and has shown to be a valuable diagnostic and prognostic tool.^23,24 Myocardial strain represents the deformation of the myocardial fibers over the cardiac cycle produced by stress and can be measured with cine-CT (Figure 1). Typically, three main spatial directions are evaluated: longitudinal, circumferential and radial strain. Global longitudinal strain (LGS) has shown to be the most consistent strain measure.²⁵ Whereas echocardiography and CMR are gold standard for strain imaging, CT has shown good correlation with 2D-echocardiography for left atrium²⁶ and left ventricle strain assessment^26–28 and moderate-good correlation with CMR.²⁹ To the best of our knowledge, no studies have been published on the incremental value of CT-derived strain for evaluation of patients with suspected or known CAD using FFR as the reference standard. In early TAVR studies, CT strain was a predictor for mortality and composite outcome after TAVR in patients with normal LVEF.³⁰ In few other clinical studies, CT-strain was evaluated in patients with myocardial infarction³¹ and congestive heart failure.³² Since CT-strain is a relatively new tool and clinically studied in only a selected group of patients, reference values are needed.^33,34 Furthermore, different techniques to measure strain have been published and their clinical impact has yet to be evaluated.^35–37

Figure 1.

Cardiac strain analysis based on multidetector row CT feature tracking (QStrain, Medis BV, Leiden, The Netherlands). The patient had suffered a lateral wall infarct. Upper left panel shows features tracked over the cardiac cycle (a). The lower left panel (b) shows the excursion of the endocardial contour over the cardiac cycle with pink denoting end-diastole and yellow denoting end-systole. Note lack of motion of the lateral wall (arrow). The line plot enables simultaneous visualization of longitudinal and circumferential strain as well as ejection fraction. A normally functioning heart would be positioned in the lower left corner of the graph. Upper right panel (c) shows the global longitudinal (EndoGLS), circumferential (EndoGCS) and radial (*GRS) strain curves over the cardiac cycle. The middle right panel (d) shows the volume of the left ventricle over the cardiac cycle (purple curve) as well its derivative (blue curve). The lower right panel shows a bullseye plot of the peak endocardial longitudinal strain per cardiac segment. See Supplementary Video 1 for a cine loop video of the beating heart. Data are courtesy of DrErasmo de la Pena-Almaguer, Tecnologico de Monterrey, Mexico.

Fractional flow reserve CT

Whereas CCTA has high diagnostic performance for ruling out epicardial CAD in low and intermediate risk patients, grading the severity of coronary stenosis and assessing hemodynamic significance based on anatomical information alone is challenging. In patients with stable chest pain, it has been shown that even severe coronary artery obstruction results in functional obstruction and myocardial ischemia in only a subset of patients.³⁸ Hence, additional assessment of hemodynamic significance of coronary artery stenosis may be required in patients with known CAD and/or stable chest pain. The standard of reference for assessment of hemodynamic significance of coronary stenosis is invasive fractional flow reserve (FFR) pressure wire measurement during invasive coronary angiography (ICA) studies. Invasive FFR is an index of the functional severity of a coronary stenosis and is defined as the ratio between the mean distal coronary pressure and the mean aortic pressure at hyperemia and represents the fraction of maximal blood flow despite the stenosis.³⁹ FFR-guided revascularization procedures have shown to reduce the rate of (MACE) events and revascularization therapies and costs.^40,41 Recent advances in computational flow dynamics (CFD) have enabled non-invasive detection of hemodynamic significant stenosis using anatomical CT data only. With CT-derived fractional flow reserve (FFR-CT), CFD techniques are applied to simulate coronary blood flow and obtain functional information.^42,43 Several elemental components are needed to run these simulations, including accurate anatomical segmentation, setting boundary conditions and modeling (micro)vascular resistance and compliance. Various FFR-CT algorithms have been investigated. HeartFlow Inc. (Redwood City, California, United States of America) was first to present commercially available FFR-CT for CAD evaluation, showing improved diagnostic accuracy in the detection of hemodynamic significant CAD in the DISCOVER-FLOW, DEFACTO and NXT trails.^44–46 HeartFlow is an offsite segmentation and analysis platform, which operates on complete three-dimensional flow simulations of the coronary artery system, requiring extreme computational power and time. Additional drawbacks are that only high-quality CCTA images without motion or step-artefacts can be used for precise anatomic modeling, in addition to costs.⁴⁷ The initial HeartFlow trials were followed by publications on FFRCT algorithms developed by Siemens Healthineers (cFFR, Siemens HealthineersGmbH, Forchheim, Germany), who employ slightly altered CFD methods allowing time-reduced, onsite evaluation using regular workstations.^48–51 Subsequently, other vendors including Canon Medical Systems (formerly Toshiba Medical Systems Corp.),^52–55 PhilipsHealthcare^56–59 and various research groups^60–70 presented onsite FFR-CT algorithms (Figure 2). A recent meta-analysis showed similar performance for offsite and onsite FFR-CT CFD algorithms.⁵ An overview of studies addressing the diagnostic accuracy of CT-derived FFR and CT perfusion is presented in Supplementary Table 1.

Figure 2.

Example of CT angiography-derived fractional flow reserve (CT FFR) analysis (FFR-CT, IntelliSpace Portal, version 9.0.1.20490; Philips Healthcare, Best, The Netherlands). In panel (a), a curved multiplanar reformation of the CT scan is shown with a 50–69% stenosis in the LAD (white arrow). In (b), a anatomic model of the coronary tree is shown with a 50–69% stenosis in the proximal left anterior descending coronary artery (white arrow) leading to a calculated FFR value of 0.75 (black arrow). In panel (c), the corresponding invasive coronary angiogram is shown, which confirmed the presence of moderate stenosis (white arrow) with a significant pressure drop distal to the stenosis resulting in an FFR value of 0.76 (black arrow).

Overall, FFR-CT showed high diagnostic accuracy for hemodynamic significant coronary stenosis, particularly when combined with CTA.^5,71 At vessel level, pooled sensitivity and specificity were 0.85 and 0.78 (FFR-CT) and 0.76 and 0.80 (FFR-CT and CTA), respectively.⁵ At patient level, pooled sensitivity and specificity were 0.89 and 0.76 (FFR-CT), considerably higher than CTA alone. Another meta-analysis presented a pooled FFR-CT sensitivity and specificity of 0.89 and 0.71 for patient, and 0.85 and 0.82 for vessel level.⁷¹ In patients with three-vessel disease, FFR-CT resulted in reclassification to a lower risk category in up to 30% of patients.⁷² It is also important to take into account the degree of coronary calcifications, as this may influence diagnostic accuracy.⁷³ In addition, it is important to consider the absolute value of the FFR-CT result, because the diagnostic accuracy of FFR-CT varies markedly across the spectrum of disease, as was shown in a recent systematic review by Cook et al.⁷⁴

At present, several trials in patients with suspected stable chest pain and intermediate degrees of stenosis have shown FFR-CT-guided treatment decisions to result in lower rates of negative ICA studies and lower costs,^75–78 with no major adverse clinical events within 90 days in patients with FFR-CT ≥0.80 in the ADVANCE registry.⁷⁸ The clinical impact of FFR-CT-guided treatment decisions in patients with multivessel disease has yet to be investigated. Preliminary results have been presented on the HeartFlow Planner, a novel FFR-CT tool that allows for geometric modeling, which could aid in predicting the expected effect of percutaneous coronary interventions in significant coronary stenosis.⁷⁹

CT myocardial perfusion

As discussed, the presence of atherosclerotic CAD does not necessarily imply functional coronary artery obstruction or myocardial ischemia. In patients with intermediate risk and/or indeterminate hemodynamic significance of coronary stenosis functional testing is indicated. Several non-invasive imaging techniques have become routine for the evaluation of patients with chronic chest pain to detect presence or absence of myocardial ischemia. As stated in recent ESC guidelines, the appropriate diagnostic test should be selected based on the clinical likelihood of obstructive CAD.⁸⁰ Possible imaging modalities include stress-echocardiography, single-photon emission CT (SPECT), PET or CMR. ICA (with or without FFR) should be reserved for patients with high event risk or for those with contradictory or inconclusive non-invasive imaging results.

Myocardial CT perfusion (CTP) is a relatively novel clinical imaging technique based on visualizing and quantifying the presence of iodinated contrast medium in the myocardium in rest and under pharmacologic stress conditions. The distribution of contrast bolus can be visualized either at a single point in time (‘static’ CTP), or, analogous to CMR perfusion, followed over time to show relative hypoperfusion in areas of ischemia or infarction. CTP images can be acquired either during one cardiac phase (static) or during multiple phases (dynamic). Most frequently, static CTP is used and an advantage of this method is the low radiation dose (3–10 mSv),⁸¹ whereas disadvantages include the need for precise contrast bolus timing, and restrictions to qualitative or semi-quantitative perfusion analysis only. An infrequently used static CTP technique to obtain perfusion information using CT is dual energy CT (DECT) perfusion. Acquisitions are generally performed with protocols similar to static perfusion. With DECT systems, image contrast can be increased by virtual monochromatic reconstructions at low keV, and likewise, image noise can be reduced at high keV reconstructions. Furthermore, reconstructed iodine maps – based on material decomposition features – can potentially be used to differentiate between normal and ischemic/infarcted myocardium (Figure 3).^82,83 The advantages and disadvantages of the different CT perfusion techniques are summarized in Table 1. The reported sensitivity and specificity of static CTP in comparison with invasive FFR ranged between 71–95% and 84–95%, respectively. For DECT perfusion, reported sensitivity and specificity in comparison with ICA ranged between 67–95% and 50–95%, respectively.⁸⁴ Dynamic CTP allows for qualitative, semi-quantitative or fully quantitative analysis with reconstruction of time attenuation curves (TAC). In contrast to static CTP, complete quantitative analysis with dynamic CTP may detect balanced or microvascular ischemia.^85,86 Reduced diagnostic image quality in presence of beam hardening artifacts can affect both static and dynamic CTP but multiphase acquisitions often provide the opportunity to select images with reduced (beam hardening) artifacts, although dynamic imaging in general is more prone to motion artifacts. Important drawbacks of dynamic CTP are not only higher radiation dose with conventional CT settings (typically between 10 and 18 mSv⁸¹) but also the more challenging acquisition protocols and complex analysis techniques.^5,84,87 In order to save dose, a resting CTP acquisition is often omitted. Although results are promising, dynamic CTP is still mostly limited to expert centers and a handful of clinical institutions in Asia.

Figure 3.

Example of spectral CT for improved detection of myocardial perfusion defect. Left panel shows conventional Hounsfield unit (HU) reconstruction in the short-axis orientation with slightly decreased HU values in the septum (black arrow). The iodine density reconstructions (b) show the perfusion defect much more clearly (black arrow) by plotting the iodine concentration in the different parts of the myocardium. Note the fivefold to ninefold lower iodine concentration in the septum. Invasive evaluation confirmed the presence of a 50–69% stenosis in the proximal left anterior descending coronary arrow (white arrow). The invasively measured FFR value was 0.72.

Table 1.

Summary of the advantages and disadvantages of different CT perfusion techniques

	Advantages	Disadvantages
Static CT perfusion	Low radiation dose	Timing of contrast bolus essential
	Clinically available – most experience	Susceptible to beam hardening artifacts
	Can be derived from same dataset as CCTA	Only qualitative or semi-quantitative perfusion analysis
Dynamic CT perfusion	Qualitative, semi-quantitative or fully quantitative analysis measurements possible	Prone to motion artifacts (caused by motion of the patient or the heart)
	Able to detect balanced or microvascular ischemia	Higher radiation dose
	Opportunity to select images with reduced (beam hardening) artifacts	More challenging acquisition protocols and complex analysis techniques
		Mostly used for research purposes
Dual-energy CT perfusion	Possibility to increase image contrast by virtual monochromatic reconstructions	Challenging acquisition protocols and complex analysis techniques
	Image noise can be reduced	Only used for research purposes
	Reconstructed iodine maps can be used to differentiate between normal and ischemic/infarcted myocardium
	Improves tissue characterization

Improves tissue characterization

CTP acquisition protocols are dependent on available CT systems, related spatial and temporal resolution and software. Static CTP can be acquired during systole or diastole, although most studies have been performed in diastole, as with regular CCTA.⁸⁴ No consensus or guideline exists stating whether rest or stress imaging should be performed first. The order of acquisition may depend on the clinical indication for functional imaging, the clinical likelihood of obstructive CAD and/or the result of coronary calcium study, typically preceding CTA/CTP. In patients without obstructive CAD on rest imaging, stress CTP could be omitted. A potential drawback of performing rest imaging first is the (often) need for administration of beta-blockers to reduce heart rates, particularly with static CTP. This may result in underestimation of hypo-perfused segments, especially when contrast medium residue from rest CTP is still present in the myocardium. Therefore, sufficient time (at least 10–20 min) should be scheduled between rest and stress acquisitions. Stress perfusion is performed by administration of pharmacologic agents (typically adenosine or regadenoson), which requires a dedicated team, proper patient selection and study preparation (e.g., ECG studies and monitoring). Scheduled delay between rest and stress imaging will also reduce the risk of hypotension induced by nitroglycerine (administered at rest CTA/CTP) and stress agents for stress CTP.⁸⁴

Multiple clinical studies have been published on the diagnostic performance of either static or dynamic CTP for the detection of hemodynamic significant CAD in comparison with SPECT, CMR or ICA. A selection of CTP studies using invasive FFR as reference test showed a pooled sensitivity and specificity of 0.81 and 0.85 on vessel level, respectively.⁵ For CTP complementary to CTA, this increased to 0.82 and 0.88, respectively. On patient level, pooled sensitivity and specificity were 0.83 and 0.79 (CTP) and 0.89 and 0.81 (CTP and CTA), respectively. In patients with coronary stents, CTP in addition to CCTA has shown to increase the diagnostic performance for evaluation of in-stent stenosis.⁸⁸ An overview of the diagnostic studies performed on CTP can be found in Supplementary Table 2. First results on the use of CTA in combination with CTP in patients with acute onset chest pain showed a reduction of ICA and revascularization rates compared with CTA alone, without significant differences for composite endpoints at 1.5 years.⁸⁹

Complementary role of functional CT over morphological assessment

Several studies have demonstrated the high negative predictive of CCTA. It can be used to safely rule out CAD in low and intermediate risk patients with stable angina and suspected chronic coronary syndrome.⁹⁰ Nevertheless, the correlation between the morphological assessment on CCTA and functional significance as assessed by invasive FFR is moderate.⁹¹ The reference standards for functional assessment according to the guidelines are stress-ECG, SPECT and PET, although none of these tests is capable of assessing all relevant functional and anatomical features simultaneously.^80,92 A combined strategy in which both anatomical as well as functional features are assessed with CT might fill this gap and improve the workflow for diagnosing CAD. If obstructive CAD is not detected by CCTA, no further functional assessment is needed, whereas functional assessment can be performed when obstructive or non-evaluable lesions are observed. In general, these strategies improve the positive predictive value without decreasing the NPV and increase the specificity while maintaining the sensitivity.^5,71,93–97 These results imply that a combined strategy improves the diagnostic workup and management of CAD. To the best of our knowledge, there are no studies that have compared combined assessment with CT-FFR and CTP to anatomical evaluation alone.

Guideline recommendations

The European Society for Cardiology (ESC) recently renewed their guideline for chronic coronary syndrome in 2019 in which CCTA has a more prominent role as first-line test to diagnose CAD in symptomatic patients (I-B recommendation).⁸⁰ Functional imaging is recommended if CCTA has shown CAD or is non-conclusive. No specific recommendations are made regarding the preferred functional tests. The National Institute for Health and Care Excellence (NICE) guideline⁹⁸ suggests, in addition to the recommendations in the ESC-guideline, that HeartFlow FFR-CT can be used as functional test since it is safe, has high diagnostic accuracy and may avoid the need for invasive diagnostics.

The role of machine learning in advancing functional CT of the heart

In the past few years, the field of machine learning (ML) has made enormous progress. Fueled by the availability of cheap graphics processing unit (GPU) hardware as well as open source programming tools, a host of studies have been published that show the potential of ML to improve assessment of CAD.

An interesting proof-of-concept study by Mannil and co-workers found that combining non-contrast CT myocardial texture analysis with ML enabled identification of patients with acute or chronic MI from controls.⁹⁹ ML can also aid in the identification of flow-limiting coronary artery stenoses. Coenen and co-workers found that a proprietary on-site workstation ML approach that applies a combination of pattern recognition and computational learning to derive FFR¹⁰⁰ improved assessment of the hemodynamic severity of coronary stenosis compared to a prior hybrid CFD approach that couples a reduced-order model for non-stenotic vessel sections with a dedicated stenosis model for the narrowed regions. Both on a per-vessel and per-patient level, the diagnostic accuracy and positive predictive value of CTA improved by adding ML-based FFR-CT.¹⁰¹ A more recent study by Li et al compared the same ML algorithm to dynamic CTP in 86 patients with stable angina and found that CTP was more accurate.¹⁰² Zreik et al used a DL approach to determine the functional significance of coronary stenosis in resting CCTA by employing deep learning analysis of the LV myocardium alone. The results demonstrate that automatic DL analysis of the LV myocardium in a single CCTA scan acquired at rest, without assessment of the anatomy of the coronary arteries, can be used to identify patients with functionally significant coronary artery stenosis with reasonable accuracy.^103,104 This approach was extended in subsequent work with an integrated DL solution for fully automated coronary artery segmentation, identification of coronary plaques and determination of the hemodynamic significance of coronary stenoses using a multitask recurrent convolutional neural network. For detection of stenosis and determination of its anatomical significance, the method achieved an accuracy of 0.80.¹⁰³ Furthermore, several studies are ongoing in which the ability of ML to improve CT myocardial perfusion analysis are being investigated.^105,106

Challenges and ongoing trials

Overall, the evidence for functional CT techniques seems promising but some challenges and limitations need to be addressed. Although cine-CT provides useful information, to the best of our knowledge no studies have been published linking CT abnormalities as detected with cine-CT strain analysis to the presence of invasively proven significant CAD. This represents an important knowledge gap and should be investigated. Clinically, however, the use of cine CT will be limited to patients who cannot undergo prospectively triggered CCTA because of irregular or high heart rates or equipment limitations.

For FFR-CT, there is a substantial body of evidence regarding its efficacy, but most of the published studies suffered from selection bias. Patients with co-morbidities were excluded from analysis as well as lesions in vein grafts or stents. This has resulted in an overestimation of the diagnostic performance and in limitations on the generalizability. Second, all studies performed so far were observational studies. No randomized controlled trials have been published comparing the conventional diagnostic workflow vs (additional) FFR–CT for clinical endpoints such as MACE. Observational FFR-CT studies, however, showed low adverse event rates in patients deferred from invasive FFR and revascularization therapy, supporting its safety to stratify patients.^75,76,78,107 Third, the diagnostic performance of FFR-CT strongly depends on the image quality (IQ). Despite most studies followed dedicated acquisition protocols, suboptimal IQ resulted in exclusion rates varying between 10 and 25%.^{44–46,78,108} The impact of IQ alterations on the diagnostic performance of FFR-CT has not been assessed. Fourth, no information is yet available on the cost-effectiveness of FFR-CT-based treatment strategies. For this, two, currently active, randomized controlled trials have been designed: the FORECAST trial (Fractional Flow Reserve Derived from Computed Tomography Coronary Angiography in the Assessment and Management of Stable Chest Pain, NCT03187639) and the PRECISE trial (Prospective Randomized Trial of the Optimal Evaluation of Cardiac Symptoms and Revascularization, NCT03702244). Both of these trials are supported by the FFR-CT vendor HeartFlow and explicitly do not include alternative methods for the determination of FFR. Therefore, these trials are inherently biased and do not represent the choices available to healthcare practitioners in clinical routine. No clinical trials assessing FFR-CT by other vendors could be identified. Clinically, application of FFR-CT will be limited to countries with established reimbursement or hospitals participating in clinical trials.

With regard to CTP, we discussed the increased radiation dose for dynamic protocols as a significant hurdle toward clinical deployment. The benefits and drawbacks of CTP from multiple different perspectives will be investigated by two randomized clinical trials: The CTP-PRO study (Impact of Stress CT Myocardial Perfusion on Downstream Resources and Prognosisin Patients With Suspected or Known Coronary Artery Disease, NCT03976921) and the PERFUSE RCT (Prospective Evaluation of Myocardial Perfusion Computed Tomography Trial, NCT02208388). 2000 intermediate and high-risk patients will be included in the CTP-PRO trial to evaluate a combined CTP and CCTA approach for suspected CAD, and the decision to perform static stress CTP or dynamic stress CTP will be based on local practice and technology.¹⁰⁶ The impact of CTP will be measured in terms of clinical decision-making, resource utilization and outcomes in a broad variety of geographic areas and patient subgroups. The PERFUSE RCT aims to evaluate the impact of CTP-guided revascularization on MACE at 1 year in 1000 patients with suspected stable CAD. Another issue that needs to be addressed when using CTP is the variability in the quantification approach (type of tracer kinetic model, hybrid approach etc.) and the variety in population used for research. These issues need to be settled before the technique will be applied in clinical practice on a routine basis.

Another important question concerns the role of ML. Rather than being used as a stand-alone technique, it is likely that ML can complement the approaches listed above and improve accuracy of CT-FFR as well as CTP approaches. This will be investigated in the ongoing CLARITY study.¹⁰⁵

Recently, the ISCHEMIA trial (International Study of Comparative Health Effectiveness with Medical and Invasive Approaches) found no reduction in major cardiac events with invasive coronary revascularization therapy compared with medical therapy in patients with stable chest pain and moderate-to-severe ischemia.⁶ The MR-INFORM trial (Myocardial Perfusion CMR vs Angiography and FFR to Guide the Management of Patients with Stable Coronary Artery Disease) showed a reduction in invasive coronary revascularization and non-inferiority to invasive FFR for major cardiac events.¹⁰⁹ Cost-effectiveness studies will provide more guidance on optimal diagnostic test combinations and treatment strategies.

Conclusions

Several cardiac CT techniques can complement analysis of coronary calcium and assessment of the presence and degree of coronary stenosis. Cardiac cine-CT, non-invasive estimation of FFR with various approaches as well as CT myocardial perfusion are all promising approaches for improved identification of patients with flow-limiting CAD. Application of machine learning will improve all of these techniques further.