About the Speaker
Camila González is a Tenure-track Assistant Professor at the Representational Intelligence for Intensive Care (RIIC) lab, Medical University of Vienna, affiliated with the Department of Anesthesia, Intensive Care Medicine and Pain Medicine and the Comprehensive Centre of AI in Medicine (CAIM). She leads a 6-year project on building an interpretable sepsis detection system for intensive care units that delivers recommendations based on medical guidelines and historical data. Previously, she was a Postdoctoral Scholar at Stanford University School of Medicine, first in the AI Development and Evaluation (AIDE) lab monitoring black-box commercial AI tools for Radiology, and before that in the Computational Neuroscience (CNS) lab forecasting the onset of alcohol use disorder from longitudinal neuroimaging data. She holds a Dr. Ing. in Computer Science from the Technical University of Darmstadt, where her dissertation on Lifelong Learning in the Clinical Open-World won the Freunde der TU Darmstadt Outstanding Scientific Achievements Award. She has received numerous honors, including the WWTF Vienna Research Groups for Young Investigators (VRG) award and the MICCAI Young Scientist Award.
Talk Abstract
Few AI systems reach the ICU bedside, despite a rapidly growing number of publications in anaesthesia and intensive care. Radiology, with hundreds of commercial products already on the market, shows the deficiencies that await us: models degrade under distribution shift, data from different sites and devices are hard to harmonize, expert labels are scarce, and prediction models remain opaque to the clinicians who must interact with them. This talk introduces developments in self-supervised representation learning, where models learn from unannotated data by predicting what they already contain, and traces the path from foundation models on electronic health records to our work at the RIIC lab: embedding multimodal ICU data into a latent space, following patients as trajectories, retrieving similar cases, identifying phenotypes and supporting treatment decisions. Throughout, the aim is systems that keep the physician in the loop, relieve their workload and work collaboratively with them.
We are looking forward to the talk!