René Herzer, Co-Founder of basebox, was a guest on the podcast "Pulse of Transformation" – together with PD Dr. Keno Bressem (TUM Klinikum / German Heart Center). A conversation about 2.5 million radiological findings, sovereign AI in hospitals and the honest question of where AI already delivers today – and where it doesn't yet.

Artificial intelligence in medicine moves between euphoria and justified skepticism. In the current episode of "Pulse of Transformation," it's all about the point where both are decided: the leap from research to clinical practice.

The guests – and a use case with substance

PD Dr. Keno Bressem conducts research at the Institute for Diagnostic and Interventional Radiology at TUM Klinikum / German Heart Center. René Herzer is building a sovereign AI platform for regulated healthcare institutions with basebox. In the conversation, both share their experiences from a groundbreaking project: structuring 2.5 million radiological finding texts with large language models.

What sounds like pure data processing is actually a key – for better research, for quality assurance and for training image-based AI models.

What the episode is about

Radiology findings as "golden labels"
Why structured findings are ideal training data – and how intelligent templates automatically choose the appropriate structure for different finding types.

Secure AI in regulated environments

How on-premises AI actually works in hospitals: with user interface, rights and role management and connection to vector databases – without sensitive data leaving the premises.

The practical hurdles
From hardware procurement to weighing cloud vs. on-premises to data protection and information security: the reality of AI implementation in healthcare, unvarnished.

The path to "AI maturity"
Why AI today mainly arrives in administrative areas – procurement, quality management, documentation, HR – and why the path to direct patient care remains demanding.

Particularly exciting: agentic AI in oncology

A highlight is the discussion about AI agents that answer complex oncological questions based on multiple documents. How do you evaluate such systems against medical assessments? And why does human validation remain indispensable?

App logic as a success factor

A central thought: How low-threshold must the interaction with AI be? The answer lies in well-thought-out app logic that makes recurring tasks – expert opinion applications, structured writing, translations – accessible to all employees.

Interesting detail from practice: "Token turnover" often doubles as soon as the first truly helpful applications are available in the organization.

Listen now

The episode offers deep insights into the practical implementation of AI in healthcare – far from marketing promises, but with concrete solution approaches and an honest assessment of today's possibilities and limitations.

👉 Here's the link to the episode

Heartfelt thanks to PD Dr. Keno Bressem and the "Pulse of Transformation" team for the insightful conversation.