Cardiac Magnetic Resonance Imaging (CMR) is a powerful diagnostic tool for assessing cardiac structure and function, providing detailed information crucial for clinical decision-making. The complexity and variability of cardiac images, however, poses challenges for accurate and efficient analysis. This PhD research aims to develop a novel deep learning framework for generalisable CMR analysis, addressing the limitations of existing methods and enhancing the clinical utility of cardiac imaging. The novel deep learning methods developed in this thesis leverage the capacity of neural networks to automatically learn hierarchical representations from raw image data. The framework is designed to advance stateof-the-art in its ability to generalise across diverse patient populations, imaging protocols, and scanner types, ensuring robust performance in real-world clinical settings. To achieve this, a large and diverse dataset of cardiac MRI scans was curated from open-source data banks, incorporating variations in anatomy, pathology, and acquisition parameters. Areas of investigation included semi-supervised learning techniques, pre-training and transfer learning, transformers and sequential networks, architectural refinements, and development of novel data augmentation strategies to mitigate respiratory artifacts and advance model generilisability. The key innovations include: (1) the development of novel synthetic label propagation techniques for precise time frame detection within the cardiac cycle to optimise the model’s performance on limited annotated data; (2) achieving well-balanced data sets with the use of synthetic labels derived from image registration of intermediate time frames, thereby fortifying the model’s adaptability to analyse scans from unseen vendors during training; (3) leveraging Pre-trained Models and MRI Specific Augmentations to mitigate respiratory artifacts and (4) synthetic balancing of CMR data sets using style transfer through deformations applied to an atlas. Notably, the proposed framework enhances predictions for previously unseen pathologies, underscoring the positive impact of these techniques across diverse settings. Collectively, these advancements represent a significant step forward in the evolution of foundational models in the realm of cardiac MR analysis.