ICBHI-CLASD Keynote Speakers

Overcoming Systemic Barriers to Build the Future of Health

The rapid acceleration of innovation in healthcare stands in stark contrast to a persistent reality: many technologies fail to achieve meaningful real-world impact. This talk explores medical and technological translation as a process shaped by structural barriers and increasing uncertainty—where managing known risks is no longer enough, and the real challenge lies in operating within open, unpredictable environments. Recognizing that innovation does not automatically lead to transformation, the session will examine both the challenges and opportunities of artificial intelligence as a decision-support tool, the reimagining of the hospital of the future as an intelligent and connected environment, and the role of Living Labs as open innovation infrastructures that enable co-creation, real-world validation, and effective adoption. Ultimately, the aim is to explore new ways of bridging research, clinical practice, and society to ensure that healthcare innovation translates into tangible, meaningful impact.

Ontologies, NLP and Foundation Models for Safer and Smarter Healthcare Systems

The growing complexity of modern healthcare environments demands AI systems that are not only powerful but semantically coherent and clinically trustworthy. This talk traces a research trajectory spanning semantic knowledge representation, natural language processing, and large language models, examining how these technologies converge to address two main challenges in health informatics: the interoperability of heterogeneous institutional knowledge and extracting actionable safety signals from unstructured clinical text. Formal knowledge representations provide the scaffolding for reliable intelligent hospital systems. This foundation is brought into dialogue with NLP-based adverse event identification, drawing on experience with medical device vigilance databases such as FDA MAUDE, where deep learning pipelines can systematically surface under-reported safety events that would otherwise remain invisible to clinical engineers and regulators. Foundation models and LLMs can reshape this landscape, offering new capabilities for semantic inference and report generation, while introducing risks that make robust prompt engineering and retrieval-augmented architectures essential rather than optional. The path toward safe and intelligent healthcare systems runs through the deliberate coupling of structured domain knowledge with modern AI, preserving interpretability and keeping the clinician meaningfully in the loop.

Building Trust in Clinical AI: Explainability and Reliability in Clinical Decision Support

TRecent advances in Machine Learning (ML) and Artificial Intelligence are reshaping decision-making in demanding domains such as healthcare, offering unprecedented opportunities for clinical support. Despite this potential, widespread adoption in clinical practice remains constrained by the inherent opacity of many ML models, often described as “black boxes”, which limits interpretability, undermines trust, and raises concerns regarding patient safety and professional accountability. This challenge is further reinforced by regulatory frameworks such as the European Union’s GDPR, which establishes the “right to explanation” and reinforces the need for transparent and justifiable algorithmic decisions. Beyond interpretability, a critical and often underexplored dimension is the reliability of model predictions in real-world clinical settings. It is not sufficient for models to produce accurate outputs. They must also provide meaningful measures of confidence that allow clinicians to assess when and how much to rely on a given prediction. In this context, pointwise reliability estimation emerges as a key mechanism to support safer and more informed clinical decision-making. This talk examines emerging approaches to enhance both explainability and reliability in ML systems, highlighting methods that move beyond post-hoc interpretation towards intrinsically transparent and uncertainty-aware models. The goal is to foster trustworthy AI systems that can be meaningfully integrated into clinical workflows, empowering healthcare professionals to use ML-driven insights with confidence and responsibility.

Classical and Modern AI for Advanced Clinical Signal Processing: Best of Both Worlds

TThis presentation explores the evolution of Artificial Intelligence in bioengineering. While modern data-driven models have gained significant attention, classical computational methods continue to offer unique advantages in reliability and interpretability. Furthermore, these classical approaches are often more efficient, requiring less computational power while effectively solving specific clinical problems. We will discuss integrating these two paradigms to achieve more robust signal processing. By combining the “best of both worlds”—the proven logic of established heuristics and the power of contemporary AI architectures—we can develop more effective, accessible, and transparent solutions for the complex challenges of digital health.