The rapid growth of heterogeneous biomedical data, spanning molecular, cellular, tissue, organ, and organism/patient levels, calls for advanced computational methods capable of integrating information across multiple biological scales and data modalities. Artificial Intelligence (AI) plays a key role in enabling such integration, supporting data-driven modelling, hybrid approaches, and the development of patient-specific digital twins. This course provides a comprehensive framework for understanding and implementing multimodal and multiscale artificial intelligence to solve complex clinical problems. Students will learn how to integrate heterogeneous data from multiple biological and clinical scales — from molecular data to patient digital twins — and how to combine data-driven, mechanistic and translational perspectives.

- Knowledge of biomedical data (e.g., medical imaging, biomedical signals, genomics, omics, clinical data) - Foundations of AI - Basic experience with scientific computing and data analysis Recommended background The following are useful, though not strictly mandatory: - introductory knowledge of deep learning - basic notions of biomedical image or signal processing

Theory: - Hierarchical nature of human biology: from molecules to systems - Data types used in biomedicine: Imaging, Physiological Signals, Omics, and Electronic Health Records - Challenges of aligning heterogeneous sources and managing multi-resolution information - Main AI fusion strategies: early, intermediate, and late fusion - Representation learning for complex datasets: Multimodal embeddings and dimensionality reduction - Integration of AI models with physical and physiological laws: hybrid modeling - Patient Digital Twins and their integration of AI into clinical workflows - Validation and robustness assessment: preventing overfitting, ensuring generalization across different hospitals, and identifying statistical biases that can compromise clinical safety Practical activity : Students will be divided into interdisciplinary groups (mixing engineers and clinicians). Each group will be assigned a specific biomedical domain where multiscale/multimodal approaches are transformative (e.g. oncology, cardiology, neurology, etc...). Each group will be asked to conduct a systematic search for state-of-the-art AI models within the assigned domain, to identify the most "pivotal" papers and to prepare a presentation with a critical analysis of their results. Assessment : Final Mini-Debate Session: Groups will present their research insights followed by a peer-to-peer discussion regarding the technical bottlenecks and clinical implications of their assigned topics.

In presenza

Presentazione orale - Sviluppo di project work in team

P.D.1-1 - Novembre