Artificial Intelligence Based Fully Automated Myocardial Function Assessment for Diagnostic and Prognostic Stratification Following Myocardial Infarction

Background: Cardiovascular magnetic resonance (CMR) imaging is considered reference methodology for cardiac functional assessment but requires manual post-processing. Whether novel artificial intelligence (AI) based automated analyses may facilitate clinical workflow and deliver similar information for risk stratification is not known. Methods: CMR data (n=1017 patients) from two myocardial infarction multi-center trials (AIDA-STEMI n=725 with ST-elevation myocardial infarction [STEMI] and TATORT-NSTEMI n= 292 with non-ST-elevation-infarction [NSTEMI]) were included. Analyses of left and right ventricular (LV and RV) volumetric parameters including ejection fraction (EF) were manually and automatically assessed using conventional and AI based software. Automated contours were visually reviewed and corrected as needed. Obtained parameters from conventional, automated and corrected automated contours entered univariate and multivariable analyses for prediction of major adverse cardiac events (MACE) defined as death, reinfarction or congestive heart failure within one-year after the acute event. Results: Both manual and uncorrected automated volumetric assessments showed similar impact on outcome on univariate (LVEF HR 0.93, [95% CI 0.91-0.95]; p < 0.001 for manual and HR 0.94 [0.92-0.96]; p < 0.001 for automated) and multivariable analyses (LVEF HR 0.95, [0.92-0.98]; p < 0.001 for manual and HR 0.95 [0.92-0.98]; p = 0.001 for automated). Manual correction of the automated contours did not lead to improved risk prediction (LVEF AUC 0.67 automated vs. 0.68 automated corrected, p = 0.49). There was excellent (intraclass correlation coefficient [ICC] 0.88 [0.77-0.93] for LVEF) to good agreement (0.7 [0.6-0.76] for RVEF) of manual and automated volumetric assessments, which was further improved by manual contour correction for LVEF and 0.74 [0.58-0.82] for RVEF). User independent volumetric analyses performed by fully automated software are feasible and results are equally predictive of MACE compared with conventional analyses in patients following acute myocardial infarction.