Expanding the role of fractional flow reserve derived from computed tomography (FFRCT) for the non-invasive imaging of patients with coronary stents: rise of the machines?

My washing machine overwhelms me with its options and its sophistication.

Uma Thurman

Coronary artery disease and multimodality imaging

Several non-invasive cardiovascular imaging modalities, such as echocardiography, cardiac magnetic resonance (CMR), positron emission tomography (PET), single-photon emission computed tomography (SPECT), and computed tomography (CT), are available for the diagnostic assessment of patients with or at risk of coronary artery disease (CAD). The choice of imaging modalities should be made by determining the best combination of tests with the least risk to the patient that are able to answer the specific clinical question posed. The concomitant presence of myocardial necrosis with myocardial ischemia, stunning, or hibernation can complicate the management of coronary lesions and often requires a multimodal imaging approach [1]. Coronary CT angiography (CCTA) has been recommended for the assessment and diagnosis of recent-onset chest pain of suspected cardiac origin. However, the high sensitivity of CCTA is typically counterbalanced by low specificity, resulting in a potential excess of invasive coronary angiographies (ICA). Further limitations are suboptimal imaging due to irregular heartbeat, calcification, and, most challenging, stent artefacts.

Fractional flow reserve derived from computed tomography

Recently, there is growing interest in CCTA-derived fractional flow reserve (FFRCT), capitalizing on three-dimensional reconstruction and computational fluid dynamics (CFD). Compared to conventional approaches, FFRCT has higher diagnostic accuracy than CCTA alone [2], and its use is associated with lower detection of lesion-free coronary arteries on invasive angiography, i.e. increased specificity [3]. Furthermore, functional assessment of coronary artery stenosis by FFRCT based on state-of-the-art computation fluid dynamics (CFD) has shown a good correlation with invasive fractional flow reserve (FFR), but it is burdened by considerable processing time and is computationally demanding.

Most recently, machine learning (ML) applications, incorporating artificial intelligence tools, have been employed to calculate virtual FFR from CT images. Indeed, ML approaches may quickly analyze large amounts of data and make systems capable of learning automatically and adapting to new inputs, typically including a combination of pattern recognition and computational learning to derive FFR. Accordingly, ML-derived FFRCT may yield diagnostic performance comparable to that obtained from CFD, as well as invasive FFR, especially in patients with de novo CAD and limited calcifications [4]. Conversely, the role of FFRCT, even obtained with ML algorithms, remains unclear in patients with prior PCI, given the challenges inherent in image acquisition and processing in stented coronary segments.

New developments after coronary stenting

In this issue of European Radiology, a carefully conducted observational study investigated the feasibility and prognostic value in predicting cardiovascular adverse events of FFRCT in patients with prior stent implantation [5]. The authors used a dedicated FFRCT software based on a deep ML platform. To validate the use of FFRCT, they retrospectively selected a cohort of 33 patients from the CHINA FFRCT study with previous coronary stent implantation and with invasive FFR assessment and CCTA images at least 3 months after the index procedure [6]. They reported that FFRCT had a good correlation with invasive FFR and an accuracy of 86% to detect hemodynamically significant in-stent restenosis. The authors then explored the role of ML-based FFRCT in predicting major adverse cardiac events (MACE) in 115 patients with stented coronary vessels and with baseline and follow-up CCTA. Statistical analysis with a supervised ML-approach (Lasso regression) and Cox proportional hazard model indicated age and follow-up ΔFFRCT/length as the only two variables independently associated with MACE at follow-up. Given the low number of in-stent restenosis ≥ 50% at CCTA, the authors however focused their model mainly on patients with low-to-moderate risk. Another limitation of this type of studies is the inability to action a post-procedure FFRCT. For example, TARGET FFR showed that targets could be identified in over 40% of patients that had undergone a PCI procedure for further treatment. If we are to identify patients at high risk of future MACE events through FFRCT, how are we to action this? Through further invasive management? Through more aggressive risk factor control? The study highlights many more such questions that pose uncertainty to the cardiovascular research community.

Where are we now?

To date, research has been intense on the feasibility and prognostic implications of FFRCT in patients with suspected CAD. This is the first valid attempt to provide an answer even in patients with prior stent implantation. Upon these premises, we can propose a tentative approach at its implementation (Fig. 1), while concomitantly searching for optimal cutoff values for FFRCT. The value of 0.75–0.80 has been validated to identify hemodynamically and prognostically significant CAD by FFR, but some studies have raised questions on the applicability of this threshold to FFRCT. Matsumura-Nakano et al [7] found a modest correlation with invasive FFR, with a significant overestimation of hemodynamic significance, and identified a 0.71–0.80 grey zone. Whether the good correlation between invasive FFR and FFRCT that emerges from this study is related to the different FFRCT protocol used (ML vs CFD), to the different target (in-stent vs native vessel lesions), or to other factors needs to be demonstrated with further studies.

Fig. 1

Clinical application of coronary CT angiography and FFRCT for management of patients who underwent coronary stent implantation. CCTA = coronary CT angiography, FFRCT = CCTA-derived fractional flow reserve, ANOCA = angina with no obstructive coronary artery disease

Where are we going?

These important insights pave the way for new applications of non-invasive coronary imaging. Interesting areas of research will be left main lesions, bifurcations, trifurcations, chronic total occlusions, and overlapping stents. Indeed, these lesions typically have a higher peri-procedural complication toll, and thus need stronger indications to motivate invasive management. Notably, left main lesions have already been studied with other imaging techniques, including intravascular ultrasound and SPECT, with favorable prognostic results [8]. Even myocardial bridges, commonly encountered with CCTA, may coexist with CAD and thus interact in the diagnostic and prognostic appraisal of patients [9]. We are hopeful that ML-based FFRCT will prove its worth even in this complex yet prevalent setting.

Rise of the machines

Is this truly the era of the rise of the machines? We are probably not there yet, but clearly new technologies applied to various areas of research will prove extremely beneficial in the short and long term, as demonstrated by the recent experience with the COVID-19 pandemic. All this will dramatically reshape cardiovascular practice and the provision of healthcare in the future for all cardiovascular patients as well as practitioners.