I'm Florian, and I work on AI for biomedical images 🧠. My path here was anything but straight: I studied psychology and sociology in Munich, worked as a freelance data scientist, and then drifted into computer science for my PhD at TUM. I never fully left psychology behind, and it still shapes how I think about experts and their perception.

After research positions at Helmholtz Munich and the University of Zürich, I'm now at the Hertie Institute for AI in Brain Health in Tübingen. There, our group, the Kofler Lab, builds bridges between the academic ivory tower and the clinic and biology lab: bringing expert knowledge into machine learning models, and machine learning models into expert workflows.

Concretely, we ask whether our metrics actually reflect what experts see. With psychophysics and eye-tracking, we study how experts judge quality, and we treat label noise and annotation uncertainty as a signal rather than a nuisance. From there, we build better tools: losses like blob loss, evaluation frameworks like panoptica, and the BrainLesion Suite for getting methods into practice. We also develop generative models, such as diffusion-based inpainting of healthy brain tissue, and work out how to judge synthetic images the way experts would, across MRI, CT, and microscopy.

Biomedical AI succeeds not by replacing experts, but by keeping them firmly in the loop.

Happy to chat about open science, reproducibility, or turning research into tools people actually use.

👉 Check out koflerlab on GitHub for our latest work.

• 🧪 koflerlab: our group's home, with all our lab repositories
• 🧠 BrainLesion: a community organization building open-source tools for brain lesion image analysis
• 🏆 BraTS Inpainting Challenge: a community benchmark we co-organize, with its own organization

All of these are collaborative efforts. Issues, pull requests, and ideas are very welcome.

📰 News about our work: I post updates on LinkedIn, and new papers land on Google Scholar.