Using Cardiac MRI to Predict Outcomes in Patients With STEMI

Overview

This prospective, multicenter observational study aims to evaluate the prognostic value of a comprehensive set of cardiac magnetic resonance (CMR) imaging parameters in patients with ST-segment elevation myocardial infarction (STEMI) undergoing primary percutaneous coronary intervention (PCI). The study integrates advanced artificial intelligence (AI) techniques to extract and analyze high-dimensional imaging features from multiple CMR sequences-including cine, strain mapping, and functional sequences-going beyond traditional measures such as infarct size or microvascular obstruction. The primary objective is to identify novel prognostic markers from routinely acquired CMR images that reflect myocardial structure, function, and mechanical deformation (strain), and to assess their association with long-term clinical outcomes. In addition to standard parameters, the study includes a detailed evaluation of left and right ventricular systolic and diastolic volumes, ejection fractions, and biventricular strain components (including longitudinal, circumferential, and radial strain), as well as left and right atrial volumes, emptying fractions, and reservoir/conduit/booster strain indices. Approximately 1000 STEMI patients will undergo CMR scanning within one week after PCI. The imaging data will be subjected to AI-based feature extraction and dimensionality reduction algorithms to uncover latent patterns associated with adverse outcomes. Patients will be followed for up to three years for the occurrence of major adverse cardiovascular events (MACE), including cardiovascular death, recurrent myocardial infarction, and heart failure hospitalization. The central hypothesis is that comprehensive CMR functional and strain-derived parameters, when analyzed using AI-driven models, offer independent and incremental prognostic value beyond conventional clinical risk factors. This study seeks to establish a data-driven, multimodal imaging framework for personalized risk stratification in STEMI patients, potentially enabling more precise post-infarction management strategies. No investigational treatment is involved. All imaging and clinical data are collected as part of routine care and analyzed retrospectively for outcome prediction.

This is a prospective, multicenter observational study designed to investigate the prognostic value of comprehensive cardiac magnetic resonance (CMR) imaging parameters in patients with ST-segment elevation myocardial infarction (STEMI) who undergo primary percutaneous coronary intervention (PCI). The study focuses on leveraging artificial intelligence (AI)-based analysis to extract predictive features from a wide range of standard and advanced CMR sequences, aiming to identify imaging-derived biomarkers that provide independent and incremental value in forecasting long-term cardiovascular outcomes. Traditional CMR indicators such as infarct size, left ventricular ejection fraction (LVEF), and microvascular obstruction (MVO) have demonstrated utility in post-MI risk stratification. However, these parameters do not fully exploit the wealth of information embedded within the full CMR dataset, especially data reflecting myocardial and atrial mechanics. In this study, we apply advanced computational methods-including radiomics, machine learning, and survival modeling techniques-to analyze multidimensional features extracted from routine, non-contrast CMR sequences. CMR image acquisition includes short-axis cine imaging and dedicated functional sequences allowing for the quantification of bi-ventricular and bi-atrial function and deformation. Specifically, the following parameters are collected and analyzed: * Left Ventricular (LV) Parameters LV End-Diastolic Volume (LVEDV) LV End-Systolic Volume (LVESV) LV Stroke Volume (LVSV) LVEF Global Longitudinal Strain (LVGLS) Global Circumferential Strain (LVGCS) Global Radial Strain (LVGRS) * Right Ventricular (RV) Parameters RVEDV RVESV RVSV RV Mass RVEF RVGLS, RVGCS, RVGRS * Left Atrial (LA) Parameters Maximum Volume (LAVmax) Pre-Atrial Contraction Volume (LAVpac) Minimum Volume (LAVmini) Total, Passive, and Booster Emptying Fractions (LAEF) Reservoir, Conduit, and Booster Strain * Right Atrial (RA) Parameters RAVmax RAVpac RAVmini Total, Passive, and Booster RAEF RA Reservoir, Conduit, and Booster Strain All images are analyzed by trained imaging specialists using semi-automated tools (e.g., CVI42) for myocardial and atrial contouring. The left ventricular myocardium is segmented at the end-diastolic phase, and regions of interest (ROIs) are exported for radiomic analysis using ITK-SNAP and PyRadiomics. Preprocessing includes voxel size resampling, normalization, and gray-level discretization. To ensure reliability and minimize redundancy, feature selection is performed in several stages. First, features with intra- and inter-observer intraclass correlation coefficients (ICC) ≥ 0.75 are retained. Second, highly collinear features are removed using correlation thresholding. Third, feature importance is assessed via random survival forests (RSF), followed by least absolute shrinkage and selection operator (LASSO) Cox regression to construct an optimized feature set. Selected features are used to calculate a radiomics-based risk score (RAD score), which is incorporated into survival models. Patients are divided into a training and validation cohort using stratified random sampling based on MACE incidence. The primary outcome is the occurrence of major adverse cardiovascular events (MACE), including cardiovascular death, reinfarction, and heart failure hospitalization. Prognostic models are developed using multivariable Cox regression and evaluated using Harrell's concordance index (C-index), calibration plots, and time-dependent receiver operating characteristic (ROC) curves. Risk stratification analyses are conducted across subgroups defined by conventional imaging markers (e.g., infarct size, LVEF, MVO). The analysis pipeline is implemented using R software, and internal validation is performed to assess model stability. Multiple imputation is used to address missing data, and sensitivity analyses are conducted to test the robustness of the predictive models under various assumptions. No experimental intervention or investigational drug is administered in this study; data collection is non-interventional and integrated into routine clinical care. The anticipated contribution of this study is to establish a multimodal, AI-enhanced imaging framework that enables individualized post-STEMI risk assessment using routinely available CMR data. By going beyond visually assessed or conventional parameters, this study may uncover novel patterns of myocardial and atrial dysfunction predictive of long-term outcomes. Furthermore, the use of non-contrast imaging sequences enhances the generalizability and safety of the proposed risk evaluation method. The protocol is approved by the institutional ethics committee. Quality control includes standardized image acquisition and analysis procedures across centers. Data processing follows pre-specified standard operating procedures (SOPs) for image segmentation, radiomic feature extraction, and modeling. Manual and automated data checks are implemented to ensure consistency and accuracy. The imaging core lab and analytic team remain blinded to outcome data during feature extraction and model construction phases. By combining cutting-edge image analysis with real-world clinical data, this study aims to inform future CMR-based guidelines for post-infarction care and to facilitate clinical translation of advanced imaging biomarkers into personalized cardiology.