Promises and challenges of AI-enabled methods for myocardial characterisation in cardiovascular magnetic resonance

Cardiac magnetic resonance (CMR) tissue characterisation is central to the diagnosis and risk stratification of myocardial disease. However, for certain techniques tissue characterisation CMR is limited by reliance on contrast agents, sensitivity to motion, prolonged acquisition times, and time- and labour-intensive image reconstruction and analysis. Artificial intelligence (AI) has emerged as a promising approach to address these challenges by enhancing and accelerating multiple stages of the CMR workflow. Deep learning methods can automate LGE segmentation, improve motion correction and image reconstruction for parametric mapping, and enable contrast-free characterisation of scar by exploiting native CMR signals, including myocardial motion and native T1 mapping. AI has also accelerated emerging techniques such as cardiac magnetic resonance fingerprinting and diffusion tensor imaging. In addition, radiomics and deep learning-based feature extraction offer the potential to derive high-dimensional tissue phenotypes and risk markers beyond those identifiable by expert clinicians. Despite these advances, translation remains limited by access to large-scale, heterogeneous training data, alongside concerns over generalisability, fairness, and interpretability, as well as barriers to regulatory approval and clinical deployment. In this mini-review, we summarise recent developments in AI-enabled myocardial tissue characterisation using CMR, highlighting both the promises and challenges for clinical translation.