Artificial intelligence–derived stress ejection fraction in stress cardiac magnetic resonance with dipyridamole: bridging past insights with future innovations

This editorial refers to ‘Artificial intelligence-based fully automated stress left ventricular ejection fraction as a prognostic marker in patients undergoing stress cardiovascular magnetic resonance’, by S. Toupin et al., https://doi.org/10.1093/ehjci/jeae168. Stress cardiac magnetic resonance (stress-CMR) has emerged as a functional test with excellent diagnostic accuracy and predictive value in patients with known and suspected coronary artery disease (CAD).1–4 However, while extensive data support the consistent and robust prognostic stratification using stress agents such as dobutamine, adenosine, and regadenoson,3,5 fewer studies have investigated the usefulness of dipyridamole6,7 in stress-CMR for predicting adverse clinical events. In addition, more than 20 years ago, nuclear medicine studies have taught us how stress left ventricular ejection (stress-LVEF) provides valuable prognostic information in patients undergoing functional tests with vasodilators,8,9 but this parameter has seldom been used in the world of stress-CMR. In this issue of the Journal, Toupin et al.10 aimed to fill the gap building on the solid foundations of the past while looking towards the future. Nine thousand seven hundred twelve consecutive patients (66 ± 15 years, 67% men) undergoing dipyridamole stress-CMR were included to determine if automated stress artificial intelligence (AI)–based left ventricular ejection fraction (stress-LVEFAI) can provide incremental prognostic value beyond traditional clinical and imaging parameters. Primary end-point was all-cause death. First, authors showed an excellent correlation between stress-LVEFAI and stress-LVEF measured by expert operators (r = 0.94, P < 0.001). This confirms previous findings regarding the high reliability of AI in assessing cardiac chamber dimensions and function.6 On top of that, Toupin et al. establish, for the first time, the value of stress-LVEF among the other parameters of the stress-CMR predictive toolkit. Indeed, after adjusting for other clinical and imaging parameters, stress-LVEFAI was independently associated with the occurrence of death in patients undergoing stress-CMR, providing additional prognostic value beyond traditional risk factors, inducible ischaemia, and late gadolinium enhancement (C-statistic improvement: 0.11; net reclassification index = 0.250, all P < 0.001). Importantly, the prognostic value of stress-LVEF remained consistent across various subgroups of clinical interest, including both women and men, as well as patients with or without known CAD, diabetes, and hypertension. In this interesting paper, Toupin et al.10 revives two past glories (dipyridamole and stress-LVEF) in a modern context through the application of an AI algorithm capable of automatically analysing resting and stress ejection fraction. The rediscovery of the importance of stress-LVEF, a parameter somewhat sidelined in stress-CMR due to the focus on identifying perfusion defects and myocardial viability, parallels the renewed interest in dipyridamole. This medication, introduced on the market as coronary vasodilator drug more than half a century ago,11 has been largely replaced in many CMR labs by adenosine and regadenoson, considered more potent vasodilators and with fewer side effects. However, the study by Toupin et al. demonstrates that stress-CMR with dipyridamole remains highly feasible and maintains an excellent safety profile. Importantly, dipyridamole offers a comprehensive approach that includes perfusion defect detection, wall motion abnormalities identification, stress-LVEF assessment, and evaluation of myocardial viability with a single exam.6 Recently, global circumferential strain, derived from feature tracking applied to stress cine imaging, has further extended the list of parameters that enable prognostic stratification of patients undergoing stress-CMR with dipyridamole.12,13 In conclusion, despite the limitations of a retrospective study, the article by Toupin et al.10 confirms the findings of extensive literature highlighting the importance of CMR as a tool for prognostic stratification in patients with known or suspected CAD. For the first time, the article describes stress-LVEFAI as an additional predictive parameter on top of other clinical and imaging standards, achievable by simply applying an algorithm of AI to stress cine imaging. If confirmed by prospective studies, this finding could lead to a revision of current stress-CMR protocols that do not include stress cine imaging when using vasodilators for calculating stress-LVEF.5 It remains to be proven whether the same results are reproducible using other vasodilators that are more commonly used today, such as adenosine and regadenoson. None declared. No new data were generated or analysed in support of this commentary.