Artificial intelligence-assisted ECG interpretation for identification of occlusive myocardial infarction following cardiac arrest: a multi-reader, multi-case study

Abstract Interpretation of post–return of spontaneous circulation (ROSC) electrocardiograms (ECGs) is challenging, yet critical for identifying occlusive myocardial infarction (OMI) requiring urgent reperfusion. While artificial intelligence (AI)–based decision-support tools show promising results, their impact on physician decision-making remains unclear. In this multi-reader, multi-case (MRMC) study​, 69 real-world post-ROSC ECGs were evaluated by 15 physicians from five groups at two timepoints (T1 [without AI] and T2 [with Queen of Hearts (QoH) neural network support]), yielding 2,070 interpretations. At T1, sensitivity/specificity were 77.97%/84.64% (LR + 5.08, LR − 0.26); at T2, 78.55%/90.43% (LR + 8.21, LR − 0.24). For the primary endpoint, MRMC AUC increased from 0.882 to 0.912 (Δ 0.030, 95% CI 0.008–0.058; p ​=0.018). The error rate fell by 2.32% (95% CI − 0.9 to + 5.6; p ​=0.156); odds of correct identification were numerically higher at T2 (aOR 1.29, 95% CI 0.98–1.69). Of 66 changed decisions, 64 (97%) followed QoH: the error rate fell by 5.03% where QoH was correct and rose by 10.56% where it was not. Sensitivity for non-STEMI high-risk OMI did not improve. Confidence increased, and within-group agreement rose through convergence on the algorithm output. The benefit of AI support is tied to algorithm accuracy.