Artificial intelligence-assisted left ventricular global longitudinal strain assessment in patients with acute myocardial infarction: a RESUS-AMI trial sub-analysis

Purpose: The aim of this sub-analysis of the RESUS-AMI trial was to evaluate the correlation of artificial intelligence (AI)-assisted echocardiographic global longitudinal strain (GLS) assessments with infarct size, left ventricular ejection fraction (LVEF) and volumes from cardiac magnetic resonance (CMR) imaging, in patients undergoing primary percutaneous coronary intervention for ST-elevation myocardial infarction. The reproducibility of GLS and other echocardiographic parameters derived with the AI-assisted software were also assessed. Methods: This is a post-hoc imaging sub-analysis of the RESUS-AMI trial. Echocardiographic LVEF, volumes and GLS were measured with AI-assisted software (CAAS Qardia 2.0) using automated and semi-automated methods. The CMR LVEF, LV dimensions and infarct size were obtained from a CMR core lab with an off-line workstation (CAAS MRV 4.1). Results: In total 169 echocardiograms were analysed and the GLS showed moderate correlation with the CMR infarct size (r = 0.58 automated and 0.64 semi-automated, both p < 0.001) and LVEF (r=-0.63 automated and − 0.65 semi-automated, both p < 0.001) from 81 CMR recordings. GLS also showed moderate correlation with the LVEF (r= -0.51 automated and − 0.67 semi-automated, both p < 0.001) from echocardiography. The inter-observer reproducibility was excellent in GLS from both the automated (intraclass correlation (ICC) = 0.94, bias = 0.08, limit of agreement (LOA) = 1.75) and semi-automated analysis (ICC = 0.93, bias=-0.68, LOA = 1.44). The intra-observer reproducibility was excellent in all echocardiographic measurements. Conclusion: GLS derived from the AI-assisted software (automated or semi-automated) could be used as a marker of LV systolic function as it correlates well the infarct size and LVEF assessed with CMR and LVEF with echocardiography.