I’m a Machine Learning Engineer and Researcher at the Medical Image Computing group of the Deutsches Krebsforschungszentrum (DKFZ) in Heidelberg, Germany, specializing in 3D radiological image analysis. Over the last two years, my work has strongly focused on large-scale self-supervised representation learning for medical imaging, alongside interactive and applied segmentation research. My work can be broadly grouped into the following areas:

• 🧠 Large-Scale Self-Supervised Representation Learning (DINOv2 / DINOv3): My main research focus over the past two years has been on DINO-based foundation models for medical imaging. Within the Human Radiome Project, I worked extensively with DINOv2, contributing to large-scale pretraining on more than 2.5 million 3D CT and MRI volumes. I currently work with DINOv3, focusing on parameter-efficient fine-tuning (PEFT) strategies across medical and non-medical image domains. Both research directions are ongoing, with publications expected throughout 2026.

• 🔧 Methodological & Interactive Deep Learning: Development of 3D medical image segmentation methods, with an emphasis on weak supervision, interactive and prompt-based segmentation, and usability-driven model design. This includes large-scale benchmarking, Napari-based interactive systems, and scalable training and inference pipelines.

• 🧠 Applied & Clinical Projects: Collaboration with clinical partners on applied radiological segmentation tasks, translating methodological advances into clinically meaningful and validated systems.

• 🛠️ Tooling & Infrastructure: Design and maintenance of open-source tools for patch-based processing, large-volume data handling, compressed storage, experiment configuration, and reproducible research workflows.

I am a daily user, contributor, and maintainer of nnU-Net, and many of my projects build upon its philosophy of strong, generalizable baselines extended toward interactive and foundation-model–based systems.

• ScribbleBench – Benchmark for scribble-supervised 3D medical image segmentation
• ParticleSeg3D – Particle-based instance-aware 3D segmentation
• M3D-CAM – Class activation mapping and interpretability for 2D/3D CNNs
• napari-sam – Interactive segmentation in Napari using Segment Anything

• patchly – Patch-based processing for very large images and volumes
• MedVol – Lightweight medical image volume reader/writer with simple usage
• confly – Configuration management for ML projects, more intuitive than Hydra
• mlarray – Fast, compressed storage for large medical or science images using Blosc2
• napari-stream – Real-time streaming of image data over the network to Napari
• tqdmp – Multiprocessing-friendly progress bars
• human-readable-id – Simple, readable, reproducible experiment identifiers