Recap of the AI developer working group held on 05/05/2025 on the topic ‘What is normal? Model-based anomaly detection for transparent and trustworthy artificial intelligence in medicine’.

🤖 What does a Twix in the supermarket fridge have to do with an epileptic lesion in the brain? How can our assumptions help us develop better AI models?

At the most recent AI developer working group on 5 May 2025, Prof. Dr Axel Mosig and Johannes Schwarz shared some fascinating insights into model-based anomaly detection and explainable AI in neuroradiology. Using SpyCAT, a spatially aware transformer, they demonstrated how even subtle epileptic lesions can be identified in brain MRIs, emphasising the importance of a deeper understanding of our presuppositions in the development of trustworthy AI models.
They also discussed the challenges of working with real-world medical data and the complexity of evaluating AI systems in sensitive medical contexts.

Thanks to our speakers for their exciting presentation and to all participants for the lively discussion! 🚀