Expert Dialogue on Digital Health & AI

Artificial intelligence in medicine frequently faces the challenge of data scarcity. Strict data protection regulations and the limited availability of large, well-annotated patient cohorts often make it difficult to gather sufficiently large datasets for training AI models that generalize reliably. One promising strategy to overcome this limitation involves the use of Foundation Models (FMs). These models are first pre-trained on vast amounts of unlabeled data to learn rich, modality-specific vector representations. They can then be fine-tuned on relatively small labeled datasets to address specific medical tasks. In this talk, we will outline the principles of FM pre-training and present practical examples of their application to challenges in medical imaging.

Data scarcity also plays a critical role in clinical research, particularly in the design and execution of clinical trials. Trials may fail due to difficulties in patient recruitment—especially in the case of rare diseases—insufficient therapeutic efficacy, or unforeseen safety concerns. To mitigate such risks, the emulation of clinical trials using real-world and synthetic data has emerged as a powerful complementary approach. This talk will highlight, through concrete examples, how state-of-the-art generative AI and causal machine learning techniques can be used to realistically emulate clinical trials and support decision-making.