Skip to content

TRUST-RAD: Trustworthy Medical AI Assistant Tools for Radiology

“TRUST-RAD” project aims to add trust layers to radiology AI assistant tools, as depicted in the figure below. To improve trustworthiness, the aim is to (i) develop models that allow the radiology AI tool to generate content that is not only precise but also highly relevant, reducing the risk of errors and hallucination, and to (ii) supplement the model’s output with trusted knowledge sources while offering transparent insights into its generation process.

This project will provide detailed specifications for a key component of the radiology AI assistant tools . It will develop a functional prototype of a trust layer with Retrieval Augmented Generation (RAG), which assists AI tools in generating content that is both accurate and highly pertinent. By augmenting and adapting the generation process, this approach mitigates the likelihood of errors and misinformation while enhancing the overall quality of the model’s output. This endeavor contributes significantly towards fostering trust in clinical settings by ensuring the reliability and relevance of generated content. RAG facilitates dynamic knowledge integration, ensuring up-to-date information. It also allows specialized knowledge integration, improving adaptability and fostering trust in clinical settings.

Integration of trust mechanisms in radiological AI assistance tools


Dr. Farhad Nooralazadeh, UZH Department of Quantitative Biomedicine

Dr. Nicolas Deperrois, UZH Department of Quantitative Biomedicine

Research Engineer <TBD>

Prof. Dr. Michael Krauthammer, UZH Department of Quantitative Biomedicine

Practice partner

Dr. med. Christian Blüthgen, Universitätsspital Zürich, Institute for Diagnostic and Interventional Radiology

Prof. Dr. med. Thomas Frauenfelder, Universitätsspital Zürich, Institute for Diagnostic and Interventional Radiology

Running time: 2024-2025

Funding in the 3rd Rapid Action Call “Digital resilience: between deep fake and cyber creativity