Cardiac Magnetic Resonance Imaging (CMR) is essential for assessing heart structure and function, but its complexity challenges analysis. This research introduces a novel deep learning framework for generalisable CMR analysis, enhancing clinical utility by improving performance across diverse patient populations, imaging protocols, and scanner types. Key innovations include synthetic label propagation for precise time frame detection, synthetic balancing of datasets using style transfer, and MRI-specific augmentations to reduce respiratory artifacts. These advancements significantly enhance the model’s ability to predict unseen pathologies, representing a major step forward in cardiac MR analysis.