Skip to main content English

CAIM Impromptu Talk by Moritz Fuchs – Ausgust 21, 2026

Diese Seite ist ausschließlich in der Sprache Englisch vorhanden.
CAIM Impromptu Talk by Moritz Fuchs –August 21, 2026

Date: August 21, 2026, 11:00

Location: Pokieser Seminar Room, AKH Vienna / Department of Radiology and Nuclear Medicine, Level 7F

Speaker: Moritz Fuchs

Title: Evaluating a Foundation Model for Intensive Care: How to unlock fast generalizable evaluation

Abstract: Prediction models for critical care are typically developed for a single task at a single institution, and degrade when deployed across hospitals with different patient populations and treatment patterns. This talk presents how we evaluated ICareFM, the first foundation model for intensive care units, at the Bosch Health Campus, examining the external evaluation work needed to determine whether such a model can generalize across both hospitals and tasks without retraining.

We harmonized data from Robert Bosch Krankenhaus with 15 other critical care datasets from ICUs and emergency departments across the United States, Europe, and Asia, comprising over 1.1 million patient stays. Through the ETH–BHC Fellowship, we validated ICareFM on in-house clinical data using a local patient equivalence (LPE) framework, which quantifies how many locally labeled patients a hospital-specific model needs to match ICareFM's performance across seven prediction tasks. At Robert Bosch Krankenhaus, staged-adapted ICareFM outperformed local models trained on over 60,000 patient stays. The talk closes by reflecting on what this evaluation teaches us about generalization claims in clinical AI, and what fast, reusable evaluation pipelines could mean for hospitals considering foundation models.

 

Bio: Moritz Fuchs is an AI Researcher and Data Scientist at Bosch Health Campus in Stuttgart, where he develops predictive models for early detection of acute kidney injury in ICU patients. He earned his PhD (Dr.-Ing.) at the Technical University of Darmstadt, researching out-of-distribution detection and uncertainty quantification in medical imaging. He previously served as PR-Officer & Vice-President of the MICCAI Student Board, received Software Campus grant (Projekt UC4S2R), and holds the ETH–BHC Fellowship.

 

This is part of the CAIM Talks series.