I came to cancer research from physics. I did my bachelor’s and master’s degrees at the Department of Physics at Heidelberg University, focusing on computational and environmental physics. My master’s thesis used a simulated ecosystem to study how parasites shape evolution. Building models of complex systems turned out to be good preparation for biology, and in 2021 I joined the Department of Applied Tumor Biology at Heidelberg University Hospital to apply machine learning to cancer research.

My first projects here brought together different kinds of patient data. As part of the Clinic 5.1 initiative, I worked on combining all of a prostate cancer patient’s medical information into a “digital twin” to support treatment decisions. I also developed methods to detect traces of Lynch syndrome, a hereditary cancer predisposition, in digitised tissue slides.

Since then my work has shifted from individual analyses towards tools the whole group can use. Today I work as an ML engineer and scientist, and I build software that takes image-based research from raw data to a usable result:

• an all-in-one toolkit that prepares whole-slide images, trains classification models and applies them, including a graphical interface for colleagues without a programming background
• a 3D segmentation tool for fluorescence microscopy that outlines cells, nuclei and micronuclei to recognise senescent cells
• models that predict molecular features such as microsatellite instability (MSI) in endometrial cancer directly from routine tissue slides

Alongside these applications, I study how reliable the underlying AI models are. In particular, I look at how simple changes to an image, such as different staining or scanning conditions, affect the way pathology foundation models “see” tissue. Next, I want to find out which genetic alterations can be read directly from standard H&E slides, and whether AI can replace expensive or invasive tests with a look at tissue we already have.

Behind the scenes, I also look after much of the group’s IT, from our compute and storage servers to this website.