Left ventricle ejection fraction assessment: artificial intelligence compared to echocardiography expert and cardiac magnetic resonance measurements

Abstract Background Cardiac magnetic resonance (CMR) is the gold standard for assessing left ventricular volumes and ejection fraction (LVEF). Artificial intelligence (AI) - based echocardiographic analysis is increasingly utilized. Purpose This study compares measurements of LVEF between echocardiography (ECHO) assessed by experts and automated AI, in comparison to CMR as the reference standard. Methods We retrospectively analyzed 118 patients who underwent both CMR and ECHO within 7 days. LVEF measured by CMR was compared with results obtained from an AI-based software which automatically analyzed all stored DICOM loops (Multi loop AI analysis). Additionally, AI results were repeated using only one best quality loop for 2 and one for 4 chamber views (One Loop AI Analysis). These results were further compared with standard ECHO analysis performed by two independent experts. Agreement was investigated using Pearson’s correlation and Bland-Altman analysis as well as Cohen’s Kappa and concordance for categorisation of LVEF into subgroups (≤30%, 31-40%, 41-50%, 51-70%; and >70%). Results Both Experts demonstrated strong inter-reader agreement (R=0.88, κ=0.77) and correlated well with CMR LVEF (Expert 1: R=0.86, κ=0.74; Expert 2: R=0.85, κ=0.68). Multi loop AI analysis correlated strongly with CMR (R=0.87, κ=0.68) and Experts (R=0.88–0.90). One Loop AI Analysis achieved superior concordance with CMR LVEF (R=0.89, κ=0.75), outperforming Multi loop AI analysis and Experts. Conclusions AI-based analysis showed similar LVEF assessment as human experts in comparison to CMR results. AI-based ECHO analysis are promising, but the obtained results should be interpreted with caution in clinical practice.Overview of the study Inter-modality concordance matrices