Association of Deep Learning-based Myocardial Infarction Size Quantification in Cardiac MRI with Cardiac Biomarker Levels.
Authors
Affiliations (3)
Affiliations (3)
- University Clinic of Radiology, Medical University of Innsbruck, Anichstraße 35, 6020 Innsbruck, Austria.
- University Clinic of Internal Medicine III, Cardiology and Angiology, Medical University of Innsbruck, Innsbruck, Austria.
- Department of Mathematics, University of Innsbruck, Innsbruck, Austria.
Abstract
Purpose To evaluate the reliability and clinical applicability of an artificial intelligence (AI)-based infarct size quantification method based on cardiac MR images in patients with ST-elevation myocardial infarction (STEMI). Materials and Methods This retrospective study included patients with acute STEMI who underwent cardiac MRI between January 2005 and October 2024. A convolutional neural network (CNN) was trained on 468 unique cardiac MRI examinations, with manual infarct segmentations serving as the reference standard. On a test set, correlations between manual and AI-determined infarct sizes and peak creatine kinase (CK) and cardiac troponin T (cTnT) levels were assessed using Pearson and Spearman correlation analyses. The predictive value of the manual and AI-based measurements for the occurrence of left ventricular adverse remodeling (LVAR) was compared using the DeLong test. Results The test set comprised 800 patients (median age, 58 years [IQR, 51-67 years]; 83% male). The CNN estimated a larger median infarct size than the manual measurements did (26.5 vs 20.1 mL; <i>P</i> < .001). The correlation with peak CK levels was greater (<i>P</i> < .001) for the automated measurements (<i>r</i> = 0.76, ρ = 0.80) than for manual segmentations (<i>r</i> = 0.68, ρ = 0.72). The same relationship was observed for peak cTnT levels (<i>r</i> = 0.66 vs <i>r</i> = 0.57; <i>P</i> = .004). Manual and AI-based measurements demonstrated comparable predictive value for LVAR (<i>P</i> = .24). Conclusion The AI-based infarct size quantification method based on cardiac MRI is comparable to manual measurement and is strongly correlated with cardiac biomarkers. <b>Keywords:</b> MR-Imaging, Cardiac, Heart, Ischemia/Infarction, Segmentation, Late Gadolinium Enhancement, ST Elevation Myocardial Infarction, Convolutional Neural Networks, Cardiac Biomarkers <i>Supplemental material is available for this article.</i> © The Author(s) 2026. Published by the Radiological Society of North America under a CC BY 4.0 license. ClinicalTrials.gov identifier: NCT04113356.