Artificial Intelligence-Enabled Prediction of Reversible Versus Irreversible Chemotherapy-Induced Myocardial Injury:...

Cancer therapy-related cardiac dysfunction remains a major cause of morbidity among cancer survivors and may interrupt life-saving oncologic therapy or lead to chronic heart failure. Conventional surveillance relies largely on serial assessment of left ventricular ejection fraction, which often detects myocardial injury only after significant and sometimes irreversible remodeling has occurred. A major unmet need in cardio-oncology is early differentiation of reversible myocardial dysfunction from permanent structural injury. Artificial intelligence can integrate multimodal clinical data, including electronic health records, echocardiographic strain, cardiac magnetic resonance tissue characterization, circulating biomarkers, genomic susceptibility profiles, and wearable-derived physiologic signals, to predict not only cardiotoxicity risk but also recovery likelihood. Early machine learning and deep learning models show improved sensitivity for identifying subclinical dysfunction compared with conventional threshold-based surveillance. Temporal trajectory modeling may further distinguish transient myocardial stunning from fibrosis-mediated cardiomyopathy, enabling earlier risk-guided interventions such as cardioprotective therapy, oncologic regimen adjustment, or intensified surveillance. Despite promising results, current evidence remains limited by single-center datasets, inadequate prospective validation, and challenges in model interpretability and clinical deployment. This narrative review summarizes contemporary artificial intelligence applications in cardio-oncology, proposes a staged framework for predicting myocardial injury reversibility, and highlights priorities for future translational research.