Cardiovascular disease (CVD) remains the leading cause of mortality worldwide, accounting for approximately one third of all deaths globally and representing one of the most significant challenges in modern medicine. Over the past decade, the rapid growth of heterogeneous clinical data sources, including cardiac imaging, surface and intracardiac electrophysiological signals, wearable monitoring data, electronic health records, and multi-omics profiles, has opened transformative opportunities for artificial intelligence (AI) analysis. Yet the full diagnostic and predictive potential of these modalities can only be unlocked through integration: cardiovascular phenomena are inherently multimodal, and no single data source captures the full complexity of disease onset, progression, and response to therapy.
This introductory talk provides an overview of the current state of the art in multimodal AI for cardiovascular prediction, framing the scientific landscape that motivates this special session. We survey active research lines spanning: (i) multimodal fusion architectures, including late, early, and cross-modal attention-based fusion of imaging, signal, and structured clinical data; (ii) representation learning and foundation models, leveraging self-supervised and contrastive pre-training across ECG, echocardiography, and cardiac imaging; (iii) federated and privacy-preserving learning, enabling multi-centre model training without raw data sharing; (iv) handling of missing and heterogeneous modalities, through imputation, masked modeling, and uncertainty-aware approaches; and (v) explainability and clinical trustworthiness, as essential prerequisites for deployment in real-world cardiology workflows.
These methodological advances are supported by a growing ecosystem of European research initiatives, reflecting a sustained commitment to AI-driven cardiovascular medicine, including CVDLINK (Horizon Europe, grant no. 101137278, 2024–2028) and CARAMEL (Horizon Europe, grant no. 101156210, 2024–2029). Their objectives, their shared methodological challenges, and open research gaps that persist across their current landscapes are examined as part of this session.