Why the study?
Optimizing the efficiency of cardiac magnetic resonance for clinical decision-making in high-risk patients with suspected obstructive CAD remains challenging, motivating the development of an interpretable multimodal prediction model integrating CMR imaging, clinical variables, and laboratory biomarkers.
Population
325 confirmed CAD patients (from 1296 with suspected CAD undergoing CMR): 239 invasive CAG and 86 CTA
Comparison
Multimodal integration model (Model 3) vs laboratory-only model (Model 1) and imaging-only model (Model 2)
Design
Prediction model development and internal validation study with cross-modality applicability-boundary testing
Answers are grounded in this paper and cite their sources.
Should not yet change CAD stratification practice; leaves open external validation and cross-modality generalizability of multimodal models.
(1) Background: Cardiac magnetic resonance (CMR) is a non-invasive modality for myocardial tissue characterization in coronary artery disease (CAD), but optimizing its efficiency for clinical decision-making in high-risk patients with suspected obstructive coronary artery disease still remains challenging. (2) Objectives: This study aimed to develop and internally validate an interpretable multimodal prediction integrating CMR imaging, clinical variables, and laboratory biomarkers for risk stratification in high-risk patients with persistent suspicion of obstructive CAD after CMR. (3) Methods: Among 1296 patients with suspected CAD who underwent CMR, 325 were confirmed to have CAD. Of these, 239 underwent invasive CAG (derivation cohort) and 86 underwent coronary computed tomography angiography (CTA, Cross-Modality Applicability-Boundary Testing cohort, deliberately designed to define the model’s applicable scope rather than conventional multicenter external validation). Three hierarchical models were developed: laboratory-only (Model 1), imaging-only (Model 2), and multimodal integration (Model 3). Performance was evaluated using area under the curve (AUC), calibration curves, decision curve analysis, and SHapley Additive exPlanations (SHAP). A point-score SHAP-visualization plot was generated to illustrate relative feature contributions of the multimodal model. (4) Results: The optimal multimodal achieved an AUC of 0.920 (95% CI 0.873–0.960), significantly outperforming laboratory-only (AUC 0.719, p < 0.001) and imaging-only models (AUC 0.888, p < 0.001). Cross-modality applicability-boundary testing in the CTA cohort (n = 86) demonstrated expected performance attenuation (AUC = 0.611), indicating that this model is intended for high-risk patients referred for invasive angiography rather than unselected CTA screening. SHAP analysis identified cardiomyopathy type, E-wave velocity, and aortic annulus diameter as top predictors. (5) Conclusions: This interpretable multimodal prediction reliably showed promising internal discrimination for obstructive CAD risk stratification but demonstrated imperfect calibration and limited cross-modality performance. Prospective multicenter external validation is mandatory before any clinical-decision-support deployment.
No takes yet. Share an insight, caveat, or question.
Wu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: