PhD defence Pedro da Costa Mateus

Keywords: Generalizable AI (Artificial Intelligence), Federated Learning, Clinical Outcome Prediction, Medical Imaging

"AI-based clinical outcome prediction with multicentric data and federated learning"

The impact of Artificial Intelligence (AI) on healthcare is undeniable becoming a crucial pillar in the ongoing pursuit of leveraging healthcare data and providing personalized care to patients. However, despite notable advancements and impressive model performance, significant challenges remain that impede the widespread integration of AI into clinical practice. Overcoming these obstacles is crucial to harnessing the full potential of AI in healthcare. This thesis explores AI’s pathway toward clinical integration through practical cases in the fields of dementia and oncology, emphasizing the development of clinically useful models while addressing some of the significant challenges that hinder clinical integration. Notably, federated learning emerges throughout this work as a promising solution to train more robust AI models by making larger datasets accessible and fostering multi-institutional collaboration. In addition, this thesis demonstrates that AI effectively extracts clinical insights from medical imaging, such as identifying aging biomarkers, while highlighting that efforts to establish reproducibility are crucial to overcoming the limited applicability and understanding currently hindering the field.

Click here for the live stream.