Abstract
Heart failure with preserved ejection fraction is a prevalent condition that carries a high morbidity and mortality, with limited treatment options. Obtaining and integrating critical mechanistic insights are essential to development of optimal treatment strategies. Machine learning (ML) has played a transformative role in advancing clinical research and can be employed to understanding myocardial dysfunction in heart failure with preserved ejection fraction. ML techniques, including supervised, unsupervised, and reinforcement learning, can be applied to cardiac imaging to identify phenotypes and extract biomarkers. Mechanistic evaluation in heart failure with preserved ejection fraction integrating advanced imaging and ML can provide information on myocardial stiffness, steatosis, and energetics. Feature extraction and feature learning techniques build upon deep convolutional neural networks, and clustering algorithms can automate detection of myocardial fibrosis, energetics, and other mechanisms. Multimodal ML frameworks, such as multifidelity physics-informed neural networks, can offer deeper insights into mechanisms, improving phenotype clustering and patient-specific interventions. This review addresses how integrating ML approaches with advanced imaging can address traditional challenges and advance precision medicine for heart failure with preserved ejection fraction, guiding targeted therapies.